Bringing AI data workers to the forefront: Aspirational débrouille strategies in Madagascar and Venezuela

Abstract

The increasing use of autonomous systems with artificial intelligence (AI) models has led to a surge in demand for data processing globally. To function effectively, these systems require vast amounts of data to be organised, labelled and cleaned. Academic and public policy literature has shown how this data work is deeply tied to the contemporary recomposition of labour institutions. The rise of a global digital labour market has been associated with the commodification of work and a rise in the constraints on workers’ autonomy. In line with recent studies on the agency of data workers and drawing on the concept of débrouille, we are interested in exploring the ways in which these workers create counter-narratives, help each other, redefine their professional identities in a positive light and regain autonomy through their daily actions. Through two case studies of Venezuelan workers on micro-work platforms and Malagasy workers in business process outsourcing (BPO), we demonstrate that data workers do not passively accept the constraints imposed by the industry. Instead, we uncover the intricate interplay between workers’ representation of the industry and their identities and strategies. We find that workers deliberately invest in industry-related skills aligned with their aspirations for a better future, often connected to entrepreneurship or migration. These skills however are often not formally recognised, allowing AI companies to benefit from them, while maintaining the illusion of annotation as unskilled labour. Any policy aimed at making the sector more ethical should take into account the local aspirations and views of data workers.

This article is a part of special theme on Datafied Development. To see a full list of all articles in this special theme, please click here: https://journals.sagepub.com/page/bds/datafied_development?pbEditor=true

Introduction

In recent years, researchers have documented the growth of outsourced IT services in different regions, such as India (Kuruvilla and Ranganathan, 2010), Brazil (Grohmann et al., 2022), Argentina (Miceli et al., 2020), the Philippines (Maquito and Andal, 2017) and several African nations (Anwar and Graham, 2020). These studies highlight the crucial role of data workers in industries where business processes have been automated or digitised. These workers maintain datasets through tasks such as categorisation, moderation and data entry.

Literature related to data work highlights the heavy dependence of AI on decentralised human labour to create training sets (Anwar and Graham, 2020; Miceli and Posada, 2022; Tubaro et al., 2020). Companies in wealthier nations, such as France and the USA, often delegate tasks to lower-wage regions, such as Madagascar and India, respectively (Gray and Suri, 2019; Le Ludec et al., 2023).

The International Labour Organisation’s (2021) research on micro-work platforms highlights how they reorganise work by globally distributing data tasks and enabling location-independent competition for better-paid assignments. This has sparked discussions about the emergence of a disembedded ‘digital global labour market’ (Graham and Anwar, 2019). Although digital labour platforms are not the only way to outsource data work – Muldoon et al. (2024) proposes a typology that underlines the importance of traditionally structured Business Process Outsourcing companies and the diversity of platform organisations – data workers are often described as particularly vulnerable to the commodification of their work, regardless of organisational setup, as they compete globally for access to data-related tasks.

Companies in the data sector portray data workers as self-employed professionals operating only at the fringes of algorithmic training, downplaying the labour-intensive nature of AI work (Irani, 2015). They also significantly impact working conditions by facilitating extensive monitoring of workers’ activities. As a result, data workers are often described as performing controlled and prescribed work that is algorithmically managed. However, recent literature (Bonini and Treré, 2024; Soriano and Cabañes, 2019) questions this portrayal by showing how they maintain their agency.

This article examines how data workers navigate constraints and exercise agency in data production chains. It focuses on the sector's local position, the representations that surround it and the ways in which workers engage with and act through these to pursue their goals. Building on previous research, this article examines the relationship between workers’ local contexts, labour organisations and their agency. It contributes to the literature by comparing two contexts where data work is embedded: BPO companies in Madagascar and micro-work platforms in Venezuela.

Using the concept of débrouille, this article demonstrates that workers have a clear understanding of the data-work sector, which they integrate into their local frameworks and personal strategies grounded in an aspirational débrouille. The article further demonstrate how the industry is transformed and adapted by these strategies, ultimately sustaining the industry's existence. By doing so, it questions the prevalence of commodification and the narrative of powerless workers.

The article is structured as follows: first, it examines the development of the sector in Madagascar and Venezuela, focusing on the local contexts in which it is embedded and their impact on the opportunities available to workers. Second, it describes the skills that workers mobilise, how they acquire them and how they utilise the spaces allowed by the contractors to develop strategies of débrouille. This article has three main contributions: it shows how data workers’ agency is locally embedded through situated strategies of débrouille; it reveals how aspirational and reflexive practices emerge within precarious data work settings; and it challenges the narrative of a unified global digital labour market by emphasising local differences.

A global labour market? The embeddedness of digital work

According to Polanyi (2001), the concept of disembeddedness arises from the commodification process, which he describes as the transformation of all social spheres, including labour, into commodities. This dynamic and political process requires limiting the influence of institutions (central banks, legislators, etc.) to enable the market to self-regulate. For Polanyi, commodification is an imposed reality on society. It requires overcoming resistance and suppressing individual agency in order to establish institutions that facilitate a total market.

The physical location of workers and their movement is closely tied to this institutional transformation that facilitates commodification. As Castles (2011) argues, local institutional contexts have been interconnected through the emergence of a ‘global labour market’, which has blended migratory flows and local economic transformation policies. He argues that the development of welfare states in Europe and the United States during the 1950s and 1960s, as well as the strengthening of labour rights through trade union actions and legislation, was closely linked to the destabilisation of agrarian economies in Asia and Africa. This led to the first waves of economic migrants to Europe, which in turn led to the commodification of migrant labour. The 1960s saw the arrival of North African rural populations in France. This provided cheap labour for the factories, leading to a period of strong economic growth.

According to this perspective, the ongoing transformation of work, particularly its digitisation, is a continuation of historical patterns. Digitalisation allows foreign workers to be integrated into the production chains of Western companies, which promotes growth and reduces labour costs. This makes it possible for these companies to employ foreign workers without relying on physical migration (what Aneesh, 2006, calls ‘virtual migration’). Wood et al. (2016) argue that platform workers rely on mutual aid networks to perform their tasks. However, they are subject to commodification, as their contractual relationships are ‘disconnected from the local institutions, leaving them exposed to fluctuations in the global online market with limited access to healthcare and social protection’. This places them in a position of strong dependency on the platforms. This viewpoint argues that platform workers are entirely disembedded, and subject to strong forces that completely strip them of their autonomy. The virtual and global nature of the marketplace further diminishes workers’ ability to resist.

However, recent developments challenge the idea that data workers are part of a unified ‘global market’. According to Tubaro (2021: 11), ‘the platform-as-market model embodied in Amazon Mechanical Turk disembeds workers, exposing them to uncertainties and lack of protections’. In contrast, the ‘multi-layer model of, among others, Pactera and Appen deeply embeds workers, with better pay but also reduced flexibility, stricter discipline and less transparency’. This suggests a complex landscape of commodification within the platform economy. Posada extends this argument by showing that Venezuelan platform workers’ ‘deep embeddedness’ creates a dual dependence on digital labour platforms and financial platforms for income (Posada, 2024). While commodification is global, he argues its effects are shaped by workers’ specific institutional, economic and social contexts. Moreover, Padios (2018) contends, that the globalisation of digital task outsourcing invokes persistent historical and institutional ties rooted in colonialism, highlighting yet another aspect of embeddedness. In the Philippines, for instance, workers’ ability to perform ‘neutral’ or ‘American’ affective labour is not incidental but rooted in a post-colonial trajectory, where language, cultural familiarity and state-sponsored outsourcing policies coalesce. These perspectives invite us to move beyond the idea of data work as a single, homogeneous space. Virtual migrants are not simply subject to a globalised market: workers’ identities, sectoral understandings, skill formation, opportunity navigation and networks of resistance and mutual aid are embedded in local institutions, social worlds and historical trajectories.

Debrouille: Putting agency beyond resistance

To go beyond the idea of a single global market in which all data workers are uniformly subjected to precarity, it is essential to examine workers' subjectivities. The literature on platform workers often raises questions about their agency in relation to the platform's algorithmic systems. Ferrari and Graham (2021) introduce the concept of ‘fissures’ to illustrate how workers navigate algorithms and pursue their own interests through everyday practices of adapting and circumventing these systems. Bonini and Treré (2024) expand on this idea, arguing that resistance and agency are effectively the same, as they both involve occupying interstitial spaces to regain freedom of movement. They highlight the ‘banal’ aspects of daily work, showing how workers navigate routine tasks ‘through and against algorithms’.

The concept that the ability to act can be a form of opposition aligns with Scott’s (2000) concept of ‘routine resistance’. In his work, he uses this term to describe how dominated peasant classes in Indonesia navigate power dynamics. Instead of engaging in direct uprisings, they adopt covert daily behaviours, maintaining external deference towards hierarchies, while creating a small sphere of autonomy. These include tactics such as retreat, wear and tear and detachment, alongside the creation of counter-narratives. This autonomous space corresponds to what Scott (2000) calls the ‘hidden transcript’ – a set of rules and discourse created exclusively for local use by peasants.

This definition emphasises the significance of acknowledging informal, individual practices. Even if not all autonomous actions can be considered oppositional, they ‘can be interpreted as oppositional acts [. . .] such practices aim to improve or mitigate their circumstances’ for those involved (Carswell and Neve, 2013: 63–64). Siles et al. (2024) build upon this understanding of agency, highlighting a form of resistance that is ‘fluid’ and non-binary, capturing the ever-present tension of the power dynamics within which it is rooted. Individuals thus mobilise a refined reading of situations, enabling them to act in their own interests, which sometimes align with those of dominant actors. Vigh’s (2009) concept of social navigation captures this fluidity by linking shifts in social formations to agents’ movements and practices within them (p. 426). It is especially useful in contexts of chronic uncertainty, such as Madagascar and Venezuela.

By embedding their concept of agency in a ‘symbiotic’ and dialectical relationship between the structure created by algorithmic affordances and the ability of agents to navigate its fissures, thereby redefining its boundaries, Bonini and Treré (2024) conceive of agency as reactive, meaning that it is determined by an externally imposed constraint. We propose a broader definition of agency, embedded in the complex network of local relations and representations. For us, no matter where in the commodification process, individuals are always socially embedded, and their actions are always motivated by their intentions. Therefore, agency involves desires or intentions, cultural projects ‘organised in and around local power relations’ (Ortner, 2006: 145). This definition of agency, which places individuals in a local circulation of representations and positions, resonates with the concept of dubria (Vigh, 2009), which transcends simple acts of survival as it ‘aims both to navigate one's way through immediate difficulties and to orient one's life positively towards the future’, and tcheb-tchib (Vium, 2016), ‘poaching strategies’ that allow individuals to appropriate and manipulate space for their survival, combining the movement of agency and creativity with formal and informal practices.

This perspective aligns with Soriano and Cabañes (2019) approach to data work in the Philippine context. They argue that workers’ perceptions relate to post-colonial imaginaries, which brings nuance to the ‘digital sweatshops’ narrative. In particular, they contend that workers derive a sense of social advancement from these imaginaries, especially those tied to a relational definition of social class. This can be harnessed and utilised within local social networks. In this view, workers’ power stems from their ability to create value from their circumstances, despite economic exploitation.

We propose inscribing our workers in this complex, dynamic mesh surrounding agency by deploying an analytical framework based on the French concept of débrouille (Casilli et al., 2023; Hugues, 2023) – a direct translation of dubria, discussed by Vigh in the Mauritanian context. We use this concept to explain the non-linear and unstable nature of labour identities. In nations where informal practices are prevalent and contradict the wage-centric definitions of the Global North, the skill of débrouille is ‘an indispensable organising talent for workers so that their activities can be adapted to an economy characterised by injustice, inequality and insecurity’. We also mobilise this concept to articulate fluid notions of identity, power, migration and coloniality across our case studies. The actors described here navigate constraints, anticipate, strategise and mobilise resources and imaginaries to position themselves within social networks and longer-term trajectories that shape their lives. By relying on débrouille-dubria, we intend to strongly underline the aspirational aspect of the agency deployed by the actors we study. This concept goes beyond other terms associated with social navigation, such as viraçao (Grohmann, 2025) or rebusque (Tubb, 2020), by introducing a temporal dimension to spatial and social (formal-informal) navigation practices.

Moreover, débrouille is broader than infra-resistance, as it encompasses how actors rely on and seize upon changes in the local socio-economic fabric, and participate in defining parts of their environment through their practices and representations. It also goes further than autonomy at work (Schwalbe, 1985) by acknowledging that the aforementioned transformations – the commodification of work, migration and the emergence of new digital professions – extend beyond productive activity and influence all aspects of society.

Indeed, the concept of débrouille captures a form of everyday agency that blends resistance, accommodation and practical ingenuity. Reed-Danahay (2007) notes that ‘débrouillardise [. . .] is used to express the ability to be resourceful, clever, or cunning in difficult situations’. Hugues (2023) builds on this, showing that some groups, such as rural agricultural workers, engage in débrouillardise as a pragmatic response to necessity. In contrast, urban working-class and lower-middle-class individuals often view it as a lifestyle choice and a means of minimising their reliance on capitalism.

While débrouille helps us understand these adaptive behaviours, it also raises a wider range of inquiries. To what extent is this agency merely reactive? And to what extent does it enable future-oriented projection? Can débrouille be understood not only as a coping mechanism while waiting for a better future but also as a form of situated aspiration – one in which workers seek to extract value from precarious conditions and reconfigure their professional identities? And how does this everyday reflexivity regarding skills, value and social positioning shape workers’ ability to endure but also to plan, learn and resist on their own terms?

These questions invite us to consider how débrouille is not only a reactive adaptation to precarity but also a proactive approach to work, through which individuals develop skills and envision themselves beyond unstable circumstances. Schwartz’s (2009) theory of invested knowledge, which posits that knowledge is rooted in personal experience and is actively constructed through interaction with situations that defy complete standardisation, aligns with this perspective. For data workers, this translates to acquiring not only technical skills (such as labelling) but also the ability to interpret ambiguous instructions, adapt to fluctuating specifications and collaborate with peers, among other competencies. To account for how this lived knowledge translates into concrete actions and positions, we adopt a definition of skills that reflects our understanding of agency. As de Terssac (2011) reminds us, ‘knowledge does not come into play automatically, experience does not speak for itself, the context is not the passive receptacle in which action takes place, action is not developed in a social vacuum’. Viewed in this light, débrouille acts as both a prerequisite and a catalyst for the acquisition of skills. It highlights the everyday forms of intelligence and adaptability that enable workers to navigate uncertainty and pursue their aspirations in challenging circumstances.

Methodology

This article is based on two independent studies carried out by members of the DiPLab research group. Both involved collecting extensive qualitative and quantitative data from platform workers, data workers, managers, business leaders and data scientists.

The TRIA project fills a gap in micro-work research by studying the Spanish-speaking world. In the course of our fieldwork, we administered 2118 surveys to workers in Latin America and Spain through the Microworkers and Clickworker platforms (2020–2021 and 2022). Additionally, in 2021, we conducted 58 semi-structured interviews with people from Microworkers (5 in Spain and 53 in Latin America), which lasted about an hour. We recruited participants by posting paid micro-task advertisements. They were compensated via the platform's payment system.

The HuSh project examines the supply chains of AI systems developed by French companies. It began with 24 interviews with managers and staff from Paris-based firms that produce AI models. This initial phase revealed a predominance of data processing outsourced to service providers in France's former colonies, particularly in Madagascar. During a second phase, 147 employees and 17 executives, managers and directors from data processing companies were interviewed. Most of the interviews took place on site over six weeks of fieldwork in Madagascar. They lasted about an hour, focusing on the workers’ tasks, career paths and professional representations. A questionnaire was also given to 294 data workers to collect information about their work conditions and demographics.

Both initiatives adopted a mixed-method questionnaire and semi-structured interview design. The conversations were meticulously transcribed and analysed using coding software based on a ‘grounded theory’ framework (Flick, 2013), beginning with data-driven categorisation and evolving into conceptually rooted classification.

This article draws on two in-depth, separate instrumental case studies (Yin, 2008), each with its own research design and questions, tailored to the specificities of its field. While this article does not aim to explain how country-specific settings shape particular organisational forms of data work, comparing these two local contexts – and how each sector is embedded in its national economy – offers valuable insights into how workers experience data labour. The two cases, which come from very different settings, were deliberately selected to show the variables and constants that shape data labour across distinct geographies.

The comparison between the two field sites was constructed through the gradual alignment of independent analytical categories. Rather than applying a predefined framework, the method involved identifying common points between the empirical axes of analysis emerging from each study, thus establishing a shared analytical ground. Since the researchers in both cases hail from the field of labour sociology, various underlying aspects were influenced by the discipline itself, including work organisation, social interactions, hierarchy, work–life balance and skillset. The use of grounded theory in both studies helped unify emerging categories within these dimensions. Consequently, both studies addressed comparable topics, such as resistance to the industry, perceptions of power, portrayals of professions and education techniques.

Background and regional context

In both Venezuela and Madagascar, platforms for micro-work (Venezuela) and BPO (Madagascar) have emerged during economic crises. This highlights the inherent movement within societies that foster the agency needed to adapt to these ever-changing situations. Despite significant differences in institutional support, data work is seen as a way to introduce foreign capital and currencies into the local economy. The key distinction is that Madagascar relies on a long-established state strategy for this capital injection, while Venezuela's approach reflects a de facto strategy to navigate a difficult economic situation.

In Madagascar: Attracting foreign investors by institutionalising the digital outsourcing sector

Madagascar's BPO industry can trace its beginnings to the 1990s, when the government expanded its investment law to attract service companies. This led to the gradual establishment of BPOs in Antananarivo (Le Ludec, 2024). However, the industry's growth has been hampered by political crises in 2002 and 2009, which weakened the local economy (Razafindrakoto et al., 2017). Madagascar's Human Development Index (HDI) stood at 132nd in 1999 and 173rd in 2021 (United Nations Human Development Programme).

Despite contributing less than 3% of the national GDP, the ICT sector remains a priority for the Malagasy government and international financial institutions, which see it as an important way to attract foreign investment and develop the emerging services industry. While the BPO sector represents only 17% of the ICT sector's overall revenue, it is experiencing rapid growth, with the number of active companies more than doubling since 2015. Despite the scarcity of data on the size of the sector and the lack of inclusion of unreported workers, Kacenelenbogen et al. (2020) and his team estimate the number of formally employed individuals in Madagascar's digital industry at 22,600, which represents 3.4% of the declared workforce on the island in 2020. It is worth noting that the majority of these positions are held by BPOs and telecommunications employees. In the BPO sector, we also estimate, based on field data, that most existing jobs are in call centres and data entry.

In a context where 95.2% of jobs are informal and mainly in subsistence agriculture, INSTAT estimates that young graduates suffer the most from unemployment (INSTAT, 2022). This shows the failure of the local job market to integrate these skilled workers. Most formal jobs are in tourism and textiles. The BPO sector is therefore the main outlet for office work.

In Venezuela: Platform work to navigate the economic crisis

Over the past decade, Venezuela has faced a severe social, economic and political crisis. Despite experiencing relative improvement, the country still maintains a high inflation rate, slightly decreasing from 305.7% in 2022 to 286% in 2023 (Observatorio Venezolano de Finanzas, n.d.). The ENCOVI survey ¹ , led by the Universidad Católica Andres Bello in Caracas, shows that household income poverty fell from 82.8% in 2023 to 73.2% in 2024. Over the same period, the employment rate dropped from 60.9% to 54.7%. Although self-employment (which includes platform workers) also declined, from 48.3% to 41.2%, it remains a significant share of the labour market.

Despite having a lower informal economy than Madagascar, Venezuela's precarious socio-economic climate obscures data on employment conditions. This is exacerbated by hyperinflation and the country's shift towards dollarisation ² . Since 2022, Venezuela's minimum wage has been fixed at 130 bolívares per month, falling from about 30 dollars in value to roughly half a dollar by 2025. Even with occasional supplements, wages remain far below living costs, with the basic food basket reaching around USD 503.73 in April 2025, exceeding the income of most workers and pensioners ³ .

In 2020, the COVID-19 pandemic exacerbated Venezuela's existing challenges, limiting job opportunities. As a result, 21% of men and 7% of women aged 18–24 chose to suspend their studies (Proyecto ENCOVI, 2022). This observation aligns with the sentiments expressed by our interviewees, primarily young people, many of whom stated that they had stopped studying and started working on platforms. The current situation may explain the increasing popularity of digital platform work in the country. According to our questionnaire, 75% of respondents use micro-work platforms as their main source of income.

Although the BPO industry in Madagascar operates within a structured labour environment, data processing in Venezuela is predominantly informal, offering no social perks. Workers find themselves vulnerable to unilateral account termination, loss of income, unfavourable customer feedback that they cannot challenge and restricted job prospects.

In both cases, externalised digital services provide access to a dynamic and globalised sector in vastly undersupplied economies. However, the different working arrangements adopted are based on infrastructure availability (internet access is expensive and unequally distributed in Madagascar, favouring workers’ concentration in office-like settings) and institutional support.

During Chávez's era, the Venezuelan government promoted technology access policies, which contributed to Venezuelans’ preference for platform work. The two programmes ‘Canaima va a la escuela’ and ‘Canaima va a la casa’ sought to improve education by distributing laptops to students and their families. By 2014, there were 2,217,065 broadband subscriptions, and about 3.5 million children and teenagers received free laptops. These measures allow Venezuelans to work on platforms, despite constant technical challenges such as power outages, slow internet and outdated computers. According to Posada (2022), the estimated number of people involved in micro-tasking activities is around 425,000, based on connections to specialised platforms (p. 67). This puts Venezuela among the world leaders in data-related work.

While distinct, the realities of Madagascar and Venezuela converge from various perspectives. Both exhibit informality and the challenges that markets face in absorbing a skilled workforce. Access to the ICT sector, propelled by a state-driven strategy in Madagascar and presented as a de facto solution in Venezuela, highlights how workers turn to the digital work sector to address their respective countries’ challenges. As explored in the following parts, they display agency despite the difficulties imposed by their personal and professional situations.

States have taken two distinct paths in terms of development and working conditions. In Madagascar, an existing sector, mainly export-oriented and tax-advantaged, favours the emergence of formal players. In contrast, in Venezuela, the deteriorating regulatory environment for exports and restricted access to the internet fosters the dominance of an underground market. In the case of Madagascar, a top-down government strategy has helped to attract foreign entrepreneurs, mainly from France. In Venezuela, the development of the sector has been much more horizontal, based on the ability of local workers to take advantage of the opportunities offered by digital work platforms.

Results: Débrouille strategies of data workers to cope with local challenges

A skilling process to face uncertainty

Both contexts illustrate distinct intentional skill development strategies shaped by the need to adapt to precarious environments. In Madagascar, BPOs leverage a skilled workforce proficient in French to serve clients from French-speaking northern countries. In Venezuela, training tends to be self-directed.

The skills described here relate to how workers mobilise a body of knowledge, allowing them to perform their tasks, maintain their position, define their identity and implement strategies. Diverse tasks and projects encourage workers to develop transferable skills, which can serve as an effective strategy for job retention, even if they are not formally recognised by employers.

Workers and clients tend to agree on a definition of a job that ‘anyone can do’. Paradoxically, both entrepreneurs and workers produce a discourse that downplays workers’ qualifications, while acknowledging that good skills are crucial for employment in this sector: a high degree of adaptability, the ability to seek information independently and the ability to follow customers’ instructions as closely as possible.

This situation primarily benefits BPOs, platforms and final clients, as they gain access to a low-cost, university-educated, linguistically skilled workforce. For BPOs, growth hinges on the qualifications of their workforce, allowing them to offer higher-value activities such as web development. Our survey in Madagascar found that 75% of workers had attended university, and many pursued further training, such as in foreign languages (33%) and IT (41%), to meet the needs of the sector. Similarly, in Venezuela, this type of work attracts a predominantly young population. The majority are under 35 and have a high level of education: 56% have a university degree and 7% hold a master's degree.

For workers, minimising the skills required for their jobs allows them to distance themselves from non-vocational positions. They often possess diplomas in unrelated fields, viewing their current role as a transitional step towards a better position. This behaviour can be seen as a form of internalising the narrative perpetuated by platforms and BPOs.

As observed by Soriano and Cabañes (2019) in their study of Filipino workers, the recognition of their skills in this situation is achieved through the invocation of a colonial mindset. Customers in both cases describe the workers as ‘docile’ and compliant, able to follow instructions to the letter. This seemingly contradictory recognition of skills allows clients to leverage the skills of skilled workers while downplaying those same skills by associating them with a colonial mentality.

It's stupid what I'm going to say, but for these jobs (…) in the end you don't have to think (…) These are processes that have to be applied by people (…) and that's what costs the most. After that it's training people to be able to do it. But once it's done, people do it as well as anywhere else, and that's the good thing (…) it's that all Madagascans are sponges for understanding (…) [and very good] at applying it! And that, it's unfortunate to say (…) but this is a country that has been really oppressed, both by [the] colonists and by the kings and queens (…)! So they've been trained more to respect things, to do their job properly. Maybe less (…) to think outside the box, which is what clients now expect from Madagascar. (French CEO, Annotation company - Madagascar)

At this juncture, the discourses of clients and workers diverge. While workers acknowledge the impact of colonial representations on their identities, they also construct an alternative narrative that frames their data worker roles as a transitional phase in their careers. They thus use these ‘colonial imaginaries’ (Soriano and Cabañes, 2019) to define a professional identity that is perceived as valuable. For example, Malagasy people see their language as insular. European languages, such as French or English, on the other hand, allow them to join the global ‘modern’ world through internet access. As a result, integration into a cutting-edge, globally oriented industry marked by office work holds great appeal. Additionally, despite workers often categorising platform tasks as low-skill, their stories reveal a wide array of acquired abilities.

[Have you picked up any new skills?]

Absolutely, yeah. When you delve into data collection, you get the hang of navigating through pages, searching for the requested information and mastering the basics of using Google Excel documents. It boils down to essentially labelling data and, you know, identifying what's there, for instance. (Venezuelan platform worker)

Far from being passive recipients of task prescriptions, many workers actively reinterpret their roles, aligning their activities with future goals, and make use of peer knowledge to resist invisibilisation. In this sense, débrouille represents a form of on-the-spot self-awareness, a way of maximising the potential of work that is otherwise undervalued.

Where debrouille becomes a critical awareness: Extracting the most value out of their work

Agency does not necessarily mean high autonomy in completing tasks, but rather the ability to build a team around their work, be useful to customers and gain transferable skills. Additionally, data workers develop strategies to ensure high-quality work and meet the client's needs, which can sometimes lead to unexpected outcomes.

Some interviewees provide examples of how they prioritised the conceptual model ‘needs’ by overcoming vague or insufficient specifications. Their strategies include disseminating knowledge about annotation tools and ambiguous data. These actions help them maintain their position. In Venezuela, these exchanges take place through self-organised virtual forums, where they exchange tricks, share experiences and protect themselves from fraudulent assignments, non-paying clients and fraudulent platforms. These communities not only improve performance and productivity but also foster skill development and social interaction, with discussions sometimes unrelated to platform work (e.g., jokes or football).

The interviewees describe how they acquire skills while working, such as enhancing digital literacy, improving writing abilities and gaining new knowledge through internet searches and YouTube videos. It reveals a strong ability to be creative, allowing them to perform tasks effectively. This underscores the tightrope data workers walk between discretion and strict adherence to guidelines. Because their tasks resist full automation, annotators must exercise judgment – not merely follow rules.

The development of differentiated strategies for acquiring marketable skills is also tied to the local social and cultural context, especially for language learning. For example, in Madagascar, data entry jobs are part of a rapidly growing export-oriented service sector, providing a major career option for many urban-qualified young people. Workers invest heavily in mastering cultural and linguistic skills as part of their individual strategies for upward mobility. This can be through specific training or through consuming French-produced or translated media through radio, TV channels and the internet. This results in high French literacy: 35% have ‘advanced’ proficiency and 38% ‘intermediate’.

With training in French there's a great future, because in Madagascar there are a lot of companies that do outsourcing. (Malagasy annotator)

Similarly, in Venezuela, platforms demand English proficiency, which presents a significant challenge that shapes workers’ learning strategies. Some use online translators, while others see it as an opportunity to develop their English skills. Venezuelans show resourcefulness in devising creative ways to manage tasks, and they value English proficiency for their career development.

Doing tasks helped me because the pages were in English, and I've expanded my vocabulary. The more I read it, the more I write it, the more I understand it. (Venezuelan computer engineering student)

Skills acquisition also reflects a significant capacity for projection outside the sector. The ability to be autonomous is also valuable in terms of individual trajectories beyond data work. Consequently, the viewpoints expressed by clients and employees regarding their professions only align to a certain extent. While clients see data workers as a docile but skilled workforce, data workers see their position as an easy gateway to social advancement. This explains their low resistance to highly constrained daily activities. While subordinated to a global economy that primarily benefits the Global North, workers in both countries incorporate skill acquisition into their survival strategies: either to improve their chances in a limited job market or to pursue entrepreneurship. This can be considered a primary consequence of the role that débrouille plays in shaping their identity.

Yet, this ability to transform exploitation into a form of capital, although limited by the current economic structure, shows how Malagasy and Venezuelan workers can navigate a difficult context, while trying to regain some of the benefits of their work. Thus, learning in this sector goes beyond mere adaptation to unfavourable conditions. It reflects the deliberate efforts of workers to use their skills to improve their socio-economic trajectories.

Annotating data or performing tasks to be connected with the labour market while waiting for something better

By joining the annotation market for Malagasy workers, or the platform work sector for Venezuelans, these individuals demonstrate their adaptability and resourcefulness. They proactively chart a course through uncertain or informal employment scenarios, with the goal of creating a more prosperous future for themselves.

The majority of Malagasy workers entered the workforce upon or shortly after finishing their studies. On average, they have had 1.74 jobs lasting at least half a year before their current position (with a standard deviation of 1.49). The majority are married (45.24%) or single (36.73%), with no children (58.99%) and 39.21% live with their parents. Respondents often describe their entry into the sector as a means of accessing the formal labour market, but not as a deliberate career choice. They did not aspire to this job nor did they train for it. Some even fear getting stuck in a sector with few progression opportunities, difficult management and low pay.

The positive points […] they pay salaries at the end of the month. For example, I worked for [COMPANY] for 2 years … but I didn't see the pay rise. […]it was the same the salary I received from the beginning of my work until the last day of my work there. (Malagasy annotator)

The majority of annotators claim to adhere to their clients’ instructions solely to maintain their employment. However, some take pride in being part of the global tech workforce. They believe that their participation in an innovative sector can enhance their future job prospects. For others, the allure of this global workforce aligns with their aspiration to ‘make a career’ by founding their own company in the same industry. However, they remain unable to do so because of low wages, poor banking access and limited social capital abroad. Furthermore, scams in the outsourcing industry reflect the racism they face, as clients often perceive them as untrustworthy just because they come from Madagascar.

[Interviewee]: We discussed that the best thing here is to be an entrepreneur rather than always working for someone, but it's not easy

[Interviewer]: Why did you stop after 3 months?

[Interviewee]: Because we weren't getting paid, after a while we thought of hiring someone but it was too risky, and then for the expenses you see, we paid the electricity, we also paid the credits (…) [At the end] we had nothing, we were like idiots, we didn't know what to do, there were the charges at the same time, we're married, we've got children…

(Malagasy Project Manager – Annotation company)

In Madagascar, data workers are overseen by a middle management layer of Malagasy team leaders and project managers. Meanwhile, strategic and client-facing roles are held by French staff. This bi-national division of labour not only creates an uneven hierarchy but also highlights the ambivalent role of local managers. They stand between top-down demands and the lived realities of workers, participating in the hierarchical control over workers’ activity. However, they can also serve as facilitators of flexibility and support, ultimately enabling workers to pursue multiple professional activities.

The advantage here is that the schedule is very flexible. We don’t have fixed hours for arriving or leaving. You can come in at noon and leave at 10 p.m., which gives me more time to do other things, to earn a bit more, like I usually do. (Malagasy annotator)

In some cases, this supportive posture goes even further. A Malagasy manager, who acts as an advisor to the company's director, addressed concerns that employees were classified as independent contractors, rather than salaried, by organising a training session on labour law.

In Venezuela, workers fully devoted to the platform aspire to return to school or acquire new skills, perhaps inspired by their data-entry experience and a constant longing for growth. For others, participating in platform jobs is a stepping stone towards launching their own business in the future, a common story driven by frustration with the government and the economic system, as well as a desire for self-sufficiency. Even then, workers often claim that this work is temporary.

In other words, I mean that even if economic activity improved [in the country], what you are asking me if I would remain [in Venezuela], no. I would still take advantage of it to generate more income [from platform job], but in the very long term I would always maintain the decision to emigrate. (Venezuelan engineer)

The longing for autonomy in the face of economic hardship has been a topic of discussion (Barchiesi, 2011). This desire can be understood as a form of resistance against what is perceived as an exploitative situation. However, limited resources often make these aspirations unattainable. This entrepreneurial approach is closely linked to migration aspirations, as both are perceived as a protective measure against exploitation, through the pursuit of better living conditions.

In Madagascar, where the industry is regulated and there are instances of entrepreneurs who have ‘successfully’ used these emerging forms of outsourcing, some employees dream of starting their own BPO company. In Venezuela, where workers are effectively independent contractors but where the informal nature of the sector and the dominance of platforms prevent them from establishing their own businesses, the only feasible course of action is to ultimately exit the industry (and the market). Workers are acutely aware of their exploitative situation and devise strategies to circumvent it, providing them with short-term relief and a glimpse of a better future. These strategies are a form of aspirational débrouille.

Clever workarounds: Circumventing rules through subtle acts of resistance

Data workers’ resistance is usually not realised through trade unions and formal worker collectives. Even though they have established labour rights, such as the right to strike, the challenging working conditions in both countries do not facilitate labour unions. However, workers develop strategies to overcome these constraints, demonstrating their resourcefulness and ability to creatively subvert rules, in a way that resists commodification.

In Madagascar, employees sometimes report that companies break local labour laws by paying piece rates for full-time positions or withholding wages in certain months. However, the prospect of severe consequences makes it almost impossible for workers to organise collectively, limiting the scope of their actions.

There were payment problems. We tried to talk to the boss, but there's always someone above the boss, so it was complicated. There was a strike, and some people were fired. (Malagasy annotator)

They also describe acts of sabotage, such as significant drops in production quality, abandoning work or taking equipment to partially compensate for unpaid wages. This form of resistance can have a substantial impact on annotation firms and AI algorithms: a whopping 93% of annotators believe that their work affects the final product's quality. In fact, clients frequently cancel agreements and look for alternative service providers when they notice a drastic drop in quality.

Venezuelan workers build solidarity through both individual and collective strategies, including ways of bypassing platform nationality quotas. When tasks appear (e.g., on YouTube), active workers alert peers to avoid missed opportunities and speed up completion. Facebook groups such as ‘Microworkers Venezuela’ help coordinate this informal cooperation and provide community and support for newcomers. Through collaboration and mutual assistance, these online communities improve the overall work experience, helping workers overcome challenges. This demonstrates the social dimension of agency (Ortner, 2006). These acts of ‘infra-resistance’ (Scott, 2000) are accompanied by counter-discourses and alternative representations, particularly in Madagascar, where feelings of exploitation are not often expressed verbally. Instead, they circulate within affinity groups of colleagues, forming a ‘hidden transcript’. This alternative set of standards, coupled with the recognition of shared knowledge, enables annotators to unite and establish a collective identity rooted in their professional experiences. Despite the form, these actions demonstrate a remarkable capacity to defy imposed norms, using informal tactics that reveal their intentionality.

My dream was to become a lawyer, because I studied law. To have a real job, a real profession. But here we use a jargon, here we call [data] work a serum, because it's work that prevents us from dying. It's about survival. Because as I said, finding work at the moment is a bit hard. (Malagasy annotator)

These counter-discourses, which circulate within affinity groups, online or in person, are particularly useful for understanding how workers in this sector represent themselves. While they may not be enough to reverse the power imbalances in data production, they do allow workers to unite around a common identity and to develop exit strategies. By producing an alternative narrative, setting up self-help networks and developing exit strategies, the workers demonstrate that they are not entirely subject to Polanyi’s (2001) total market. Their ability to free themselves from the data market and improve their position demonstrates some resistance to the commodification of their work.

Conclusions

Our contribution highlights the embeddedness of data work in its own unique economic and social networks. Tubaro (2021) underscores the complex web of inter-firm relationships that enable this type of work, while Posada (2022) emphasises the critical networks of mutual aid that are essential for its operation. This understanding can be extended by arguing that this work is also inscribed in strategies and workers’ ability to débrouille. Therefore, data work becomes part of a larger process for annotators to redefine their identities, aligning with their migration plans, their position in the industry and their perspective on it.

The results show local differences, particularly regarding data work organisation. These disparities affect the emergence and operationalisation of débrouille strategies. In Madagascar, for example, BPOs follow a hierarchical structure based on salaried employment. Workers also receive structured training both before and during annotation projects. In micro-work platforms in Venezuela, independent contractors independently select tasks at a piece rate, often without any structured training. In Madagascar, the existence of a formal sector allows some workers to project themselves as future entrepreneurs, including through attempts to establish their own annotation companies. In contrast, in Venezuela, informality combined with the structure of micro-work platforms – which involves a tripartite relationship across three distinct geographical locations – does not enable the same kind of projection. In Madagascar, where annotators share a physical workspace, they have the opportunity to engage in immediate forms of resistance, such as the strike attempt described in the results, although such actions remain uncommon. In Venezuela, where workers are physically separated, they rely on social media to organise and bypass platform-imposed rules, such as quotas. Middle managers, who are responsible for training, serve as a direct point of contact in Madagascar, allowing annotators to ask questions about their tasks. This reduces the need for workers to consult online forums. Middle managers occupy a complex role, simultaneously enforcing control over workers while also serving as valuable resources beyond their training responsibilities. One manager even organised an informal session on workers’ rights, as mentioned in the results.

In conjunction, our findings and the contrasting settings discussed in the background section implicitly emphasise the crucial role of the state in shaping local data industries. They not only bring to light institutional disparities but also reveal how policy and regulation impact resource access, skill development and governance infrastructure. Although unpacking these mechanisms in detail lies beyond the scope of the present study, it represents a crucial research challenge. It is essential for understanding and effectively supporting the sustainable development of the sector. Therefore, we intend to explore this dimension more fully in future work.

Our results also show that, through situated practices of débrouille, workers actively shape the conditions under which their labour is exchanged, challenging the idea of a unified, placeless ‘global labour market’. In reality, this market is embedded in locally situated strategies, circulating discourses and specific representations. Despite these differences, certain characteristics appear to be common across the data work sector. In both cases, workers described their jobs as ‘anyone could do’, despite having university backgrounds and showing specific skills. This reflects the transient nature of data work, which is often used as a stepping stone to better opportunities. In this context, débrouille consists of how workers navigate or get around constraints to achieve their goals. Débrouille is thus largely shaped by necessity, as workers face structural challenges: a labour market that fails to absorb young graduates; widespread difficulty in securing stable employment, both financially and contractually; and inflation that undermines the value of primary income. In this context, workers’ strategies align with what Hugues (2023) describes as a form of débrouille ‘out of necessity’ that is detached from any overt political meaning. This need is particularly evident in workers’ tendency to hold multiple jobs. In Venezuela, platform work is often added to a primary occupation. In Madagascar, data work is treated as the main job, but is frequently supplemented by other activities.

Our observations provide avenues for expanding the notion of débrouille, moving beyond its association with survival and exploring its role in forward-looking strategies. These strategies hinge on a network of workers, through which they circulate counter-discourses that allow them to carve out a space of autonomy. Workers demonstrate resourcefulness by proactively seeking ‘exit strategies’, fostering self-employment and championing ‘desertion’ from their subordinate status. This behaviour is consistent with the findings of Soriano and Cabañes (2020) in the Philippines. It's important to note that migration in this context goes beyond just a way for workers to escape poverty and insecurity in their home countries. Instead, it involves a set of carefully crafted strategies for skill acquisition and career advancement, linked to local representations of Western countries. In Madagascar, workers mobilise what Soriano and Cabañes (2019) call ‘colonial imaginaries’ to define a professional identity perceived as valued and modern. This includes the use of European languages as a means of accessing global modernity through the internet. In Venezuela, many see these tasks as temporary solutions – activities that might later be revalued or mobilised when they can resume formal education. As a result, acquiring new skills has become crucial for their survival strategies. This includes pursuing career advancement outside the industry or even considering entrepreneurship.

Therefore, débrouille relies on the reflexivity of workers to take advantage of their professional situation. Malagasy individuals aspiring to launch their own ventures often tap into the trust they have cultivated with existing clients, using it as a springboard to help them ‘get started’. This proactive step towards self-improvement, learning new skills and valuing their existing expertise is partially unconscious. Their own belief that the work is so simple that anyone could do it often undermines it. Despite this ambivalence, however, their strategies lead to the acquisition of valuable skills. In this sense, débrouille actively shapes their identity through strategies that help them make the most of the present, while also hoping for something better.

Waiting is recast as a proactive act, and activities such as learning, waiting or transitioning between tasks are re-evaluated. This fosters an environment conducive to the emergence of alternative viewpoints or counter-discourses. Although confronting institutional or political resistance can be challenging, these counter-discourses empower workers to employ various ‘exit strategies’ and leave the sector when work arrangements are unfavourable. These shared representations contradict the ‘enchanted’ narratives that platforms, BPOs and start-ups promote about data processing and AI.

In this sense, in this article, we refine a specific aspect of social navigation, and we introduce a concept that we call aspirational débrouille. Workers use their imagination to define a professional identity that includes its own exit strategies. This identity is oriented towards achieving something better and relies on specific representations of the globalised economy; workers aspire to participate in international fluxes. Some do so by migrating, while others seek to gain access to international financial or social capital that can help propel their own businesses. However, the way they implement these strategies is strongly influenced by their local institutional and social environment. As a result, what circulates in these global fluxes is not commodified ‘micro-tasks’ but resources and representations. Therefore, data work for AI is embedded in complex, intertwined global networks, rather than in a ‘total market’ as defined by Polanyi. However, access to these global fluxes is not evenly distributed among Venezuelan or Malagasy populations. In our studies, the individuals involved in data work are young, urban and hold a university education, which are all essential requirements for engaging with contemporary international flows. The form of aspirational débrouille that we identify, then, concerns a very specific social class in the countries we study.

Technological companies benefit from these fragmented, aspirational spaces – enabled by workers. This shift in débrouille, from necessity to projection, opens up limited and fragile opportunities for resistance and transformation. It is precisely this hope-driven dynamic that allows companies to pilot low-cost proof-of-concepts, undertake underfunded projects and structure the sector around short-term, back-and-forth outsourcing arrangements (Le Ludec et al., 2023), including by shaping data work as temporary. The temporary nature of this work is seldom called into question by professionals involved in AI creation, especially model designers, who persistently see automation as a process that will ultimately replace human labour. However, although the AI industry is constantly evolving, this does not necessarily mean that data work is decreasing. As models become more advanced, they require more precise, context-specific annotations and ongoing verification. The increasing complexity of AI and the persistent failures of automation actually reinforce the need for human-in-the-loop interventions. This leads to a long-term demand for more specialised human labour in tasks such as prompt refinement, edge-case classification and quality assurance (Laux et al., 2023; Rädsch et al., 2024). As we argued in a previous contribution, annotators have to develop domain-specific expertise to handle data (Le Ludec et al., 2023), which contributes to the consolidation of this professional group and to a growing organisational complexity of annotation work, with the emergence of new annotation-related roles and occupations describe in the case of Madagascar (Cornet, 2025; Le Ludec, 2024). From this perspective, workers are actively reshaping the industry by demonstrating their expertise and bringing visibility to the centrality of quality (and quality assurance) as a key issue. Yet annotation is not recognised as a profession, although it plays a crucial role in the development of AI.

This paradox leads to a broader line of inquiry. AI can be seen as a débrouille economy, one that functions through ongoing back-and-forth adjustments between different layers of the production chain. In such a configuration, AI systems rely almost entirely on workers’ own strategies of débrouille to absorb uncertainty, fill operational gaps and ensure continuity. This opens up a second avenue for exploring the negative dimensions of débrouille. Indeed, the concept also resonates with what Pettit (2024) describes in the Egyptian context as the labour of hope – ‘hope-filled practices’ such as training, migration or entrepreneurship that sustain aspirations for a better life even when structural conditions remain unchanged. According to Pettit, these behaviours ‘produce temporary distraction from difficult emotions and sustain a vital sense of hope for future mobility’ (p. 5). This demonstrates how capitalism harnesses and channels the aspirations of those it simultaneously endangers and serves to nuance the positive aspects of débrouille. In both Madagascar and Venezuela, workers’ aspirations and abilities become assets that can be traded on data work markets. Their skills, which they developed in pursuit of a better life, are used to boost the value of AI companies.

Finally, subtle forms of resistance used by annotators are unlikely to spark a complete overhaul of the global economic system. Technological corporations and start-ups predominantly reap the benefits of production, while workers remain at the mercy of the volatile data labour market, being subject to the commodification of their work. However, these strategies allow them to carve out a space of freedom and resist being passive subjects of globalisation ⁴ .

Understanding those localised strategies for débrouille is of great importance, since data work and micro-work regularly appear in international reports as a strategic axis for development (Kuek et al., 2015). Such recommendations, which promote a top-down, uniform approach, overlook critical aspects that we highlight in our work. Although these workers are highly skilled, they face structural obstacles that prevent them from moving up the value chain, such as starting their own businesses. To solely focus on micro-work or development aid as a way to integrate a global formal economy without addressing local barriers to workers’ needs is a Western-centric solution that silences workers’ voices. It is crucial to acknowledge the strategies of débrouille in order to challenge the perception that data work for AI is merely transitional, a temporary solution before full automation. Through the débrouille practices shown by workers, we demonstrate that, far from being unskilled labour ‘anyone can do’, data work is grounded in a set of under-recognised skills. Even referring to ‘colonial skills’ may unintentionally diminish the skills of these workers. Understanding how workers use their skills and resources to improve their future prospects can offer a more realistic view of the industry's role in development.

This goes beyond its macroeconomic impact, and instead looks at workers’ individual and collective professional trajectories. Furthermore, unpacking the concept of débrouille along the dimensions of necessity, future-orientation and reflexivity in work practices brings to light skill-development processes that are not yet fully theorised in the literature. This analysis points to concrete avenues for development and support aimed at professionalising and sustainably nurturing the sector. Here, we suggest that there is room to reimagine development pathways for the sector. This would involve reframing data work as interpretative and knowledge-intensive, rather than as temporary and invisible. This would mean recognising the workers’ intimate understanding of data, their ability to identify edge cases, and, in some cases, their contribution to the maintenance and refinement of AI models. As such, promoting the professionalisation of data annotation could lead to more equitable, contextualised forms of development and fairer value distribution. This would align with the needs of workers and the evolving requirements of AI.