Whether you're looking for an angel investor, a growth advisor, or just want to connect — I'm always open to great ideas.
Get in TouchAI, startups & growth insights. No spam.
TL;DR: DoorDash launched a standalone "Tasks" app on March 19, 2026, paying its network of gig workers to record Spanish conversations, film household chores, and photograph restaurant interiors — all to generate training data for AI models. The app excludes workers in California, New York City, Seattle, and Colorado, citing the regulatory environment in those regions.
The company that brought burritos to your door now wants something different from its army of gig workers: your voice, your home, and your neighborhood — all packaged as training data for the next generation of artificial intelligence.
On March 19, 2026, DoorDash quietly rolled out a standalone mobile application called Tasks — a product that has very little to do with delivering food and everything to do with feeding machine learning pipelines. The app is distinct from the main Dasher platform and is available as a separate download, though it leverages the same identity infrastructure and onboarding relationships DoorDash has built with its existing contractor base.
The premise is simple: workers browse a marketplace of short, structured tasks, accept ones they want to complete, perform the required data collection activity in the physical world or via their smartphone, submit their output, and receive payment after a review window. Tasks are listed with upfront pricing before a worker commits — an explicit design choice that signals DoorDash has absorbed lessons from the wider gig economy about transparency and worker trust.
The app's interface reportedly mirrors the familiarity of the core Dasher experience, with a task feed sorted by proximity, payout, and estimated completion time. Workers can filter by task type and choose to stack multiple assignments in a single outing, much like batching deliveries. According to reporting from TechCrunch, the rollout was initially soft-launched to a subset of existing Dashers before being opened more broadly on launch day.
Payment processing takes one to three business days — a model borrowed directly from DoorDash's existing contractor payment cadence. There is no minimum payout threshold noted at launch, though DoorDash reportedly reserves the right to reject submissions that do not meet quality standards, in which case payment would not be issued for the flagged task.
The strategic logic is compelling. DoorDash already maintains a contractor network numbering in the hundreds of thousands across the United States. Reactivating that network for data collection — without the logistical overhead of food delivery — creates an asset that is simultaneously cheaper to deploy and potentially more valuable per hour to the company than another sandwich run.
The initial task catalog at launch focuses on three primary categories, each targeting a distinct AI development use case.
The first and most notable is conversational audio recording in Spanish. Workers are prompted to speak specific phrases, sentences, or short dialogues into their phone's microphone according to provided scripts. This type of data — natural language audio from diverse speakers in varied acoustic environments — is foundational for training speech recognition systems, voice assistants, and real-time translation tools. Spanish is among the most underrepresented major languages in high-quality voice datasets relative to its speaker population, making it a priority for companies building multilingual AI products. The demand here is substantial: a single production-grade voice model may require millions of spoken utterances across hundreds of speaker profiles.
The second category involves filming household chores and everyday domestic activities. Workers point their camera at tasks like folding laundry, loading a dishwasher, wiping a counter, or organizing pantry shelves. This video data feeds into computer vision models being trained to understand physical environments, object manipulation, and human activity sequences. These capabilities underpin a range of emerging AI products, from robotic household assistants to insurance claim automation systems that need to assess property condition from visual evidence.
The third category — photographing restaurant interiors, menus, storefronts, and food preparation areas — is the most obviously aligned with DoorDash's core business. AI systems used in restaurant discovery, menu digitization, health inspection automation, and logistics optimization all require large, labeled datasets of real-world food service environments. DoorDash's existing presence in tens of thousands of restaurant partnerships makes its worker base uniquely positioned to collect this data at scale and with minimal friction.
Additional task types are expected to expand over time. VentureBeat has noted the broader industry trend toward diverse multimodal datasets, and DoorDash's infrastructure positions it to add image annotation, sentiment labeling, or even neighborhood mapping tasks in future iterations.
Unlike many crowdsourced data labeling platforms that obscure per-task rates until after a worker has committed time, DoorDash's Tasks app displays upfront pricing for each assignment before acceptance. This transparency is a meaningful differentiator and aligns with growing regulatory pressure — and worker advocacy — around clarity in gig compensation.
Task payouts vary based on complexity, time investment, and the sensitivity of the data being collected. Simple photo documentation tasks — snapping a few exterior shots of a restaurant — sit at the lower end of the payout spectrum. More involved assignments, such as multi-segment audio recordings requiring pronunciation accuracy and retakes, command higher rates. Video documentation of household activities, which requires both filming and potentially some basic metadata tagging, falls somewhere in between.
DoorDash has not published a fixed rate card for task categories, which gives the company flexibility to adjust pricing based on supply and demand dynamics within its marketplace. If a particular type of training data becomes scarce or highly requested by buyers, the platform can presumably surface higher-paying tasks to attract completions. This is a standard mechanism in crowdsourced labor markets — Bloomberg has covered similar dynamic pricing models at other data annotation platforms — but it also means workers have limited ability to predict income from the platform.
Payment is processed within one to three business days following submission and quality review. If a submission is flagged as non-compliant — out of focus, incorrect language, mismatched activity — the worker may not receive compensation for that task. DoorDash has not publicly detailed its dispute resolution process for rejected submissions, which represents an open question for workers relying on the platform as a meaningful income source.
The Tasks app is explicitly unavailable to workers in California, New York City, Seattle, and Colorado. DoorDash has not issued a detailed public statement explaining these exclusions, but the regulatory context in each jurisdiction makes the rationale transparent to anyone familiar with the current gig labor legal landscape.
California's AB5 and subsequent Proposition 22 framework created an ongoing legal tension between gig companies and the state around worker classification. Any new product that deepens the economic relationship between DoorDash and its contractors — particularly one that involves collecting biometric-adjacent data like voice recordings — risks drawing fresh scrutiny from California's Labor Commissioner or Attorney General. The state's Biometric Information Privacy protections, while less prescriptive than Illinois' BIPA, are evolving, and the California Privacy Rights Act adds additional layering around audio and video data collection from individuals.
New York City's Local Law 144 and broader package of gig worker protections, including mandatory pay minimums for app-based workers, would complicate the economics of per-task pricing if Tasks workers were classified as subject to those floors. Seattle has enacted some of the most aggressive gig worker compensation ordinances in the country, creating a similarly challenging environment.
Colorado's CARES Act successor legislation and evolving AI-specific data privacy rules — including provisions around automated decision-making and data collection consent — create meaningful compliance overhead that DoorDash has apparently determined is not worth absorbing at launch.
The practical consequence is that the regions with the largest concentrations of DoorDash workers, and the regions where worker advocacy organizations are most active, are precisely the ones excluded from the program. CNBC has reported on how gig platform companies increasingly route new products around regulatory hotspots in their early rollout phases before building compliance infrastructure for expansion.
DoorDash has confirmed that training data collected through the Tasks app will be used both internally and sold to partner companies. The internal use case is the more straightforward of the two: DoorDash is actively developing AI capabilities across its logistics platform, restaurant recommendation engine, and customer service infrastructure. High-quality labeled data from its own worker network — particularly restaurant and food service imagery — is directly applicable to these internal model training pipelines.
The partner sales dimension is where the scope of the program becomes more interesting. DoorDash has identified retail, insurance, and hospitality as the primary verticals for third-party data licensing. This is not surprising given the task types at launch. Retail companies building inventory management AI need shelf and product imagery datasets. Insurance companies developing automated claims processing need household environment video. Hospitality chains refining their digital presence need consistent, high-quality storefront and interior photography at scale.
The economics of this are substantial. Training data for specialized AI applications commands significant prices — the market for high-quality labeled datasets is projected to reach tens of billions of dollars annually as model development accelerates. By inserting itself as a data broker between its contractor workforce and corporate AI buyers, DoorDash is creating a revenue stream that has no marginal cost of delivery, no food spoilage risk, and no restaurant commission structure to navigate.
Workers, notably, do not appear to receive any revenue share from the downstream licensing of their submissions. The payout for the task is the entirety of their compensation, regardless of how many times DoorDash's platform resells or repurposes the resulting dataset.
DoorDash's Tasks launch is not happening in isolation. It is the most visible corporate manifestation of a structural shift that has been building in the gig economy for several years: the realization that a distributed, always-on, geographically diverse human workforce is one of the most efficient data collection instruments ever assembled.
Uber launched similar experiments with its driver network. Amazon's Mechanical Turk has long served as the canonical crowdsourced labeling platform, though its scale and reputation have both suffered from persistent concerns about wage suppression. Platforms like Remotasks and Toloka built their entire business models around this concept. What DoorDash brings that these predecessors lacked is a pre-vetted, identity-verified worker pool with physical access to the real-world environments that AI training pipelines increasingly need: homes, restaurants, public spaces, vehicles.
The evolution from "deliver this burrito" to "record this sentence" is smaller than it appears. Both tasks ask a contractor to use their body, their time, their equipment, and their presence in specific places to generate value for the platform. The difference is that data collection tasks have no routing complexity, no customer interaction requirement, no temperature sensitivity, and no two-sided marketplace coordination cost. From a pure operational perspective, data tasks are dramatically cheaper to orchestrate than delivery tasks.
Wired has chronicled the broader arc of how AI development has quietly become the gig economy's largest emerging employer, with millions of workers globally performing data annotation, content moderation, and model evaluation work that rarely surfaces in public AI product narratives. DoorDash's Tasks app brings that dynamic into explicit view for a mainstream audience.
The launch of Tasks raises a set of ethical questions that DoorDash has not yet addressed in any substantive public communication.
On consent and data rights: workers who record their voices, film their homes, or photograph their neighborhoods are generating data that will be used to train AI systems for commercial gain. The legal consent framework — a terms of service acceptance — is standard but thin. Workers may not fully understand that a recording of their voice reading a Spanish phrase could become part of a permanent training dataset used across multiple commercial AI products for years. There is currently no mechanism for workers to request deletion of their specific contributions from a training dataset, nor any clarity on whether such deletion would even be technically feasible once the data has been incorporated into model weights.
On privacy: the household chore video category is particularly sensitive. Workers filming their own homes are, by necessity, capturing images of their personal living spaces, potentially including identifying information, household members, or evidence of economic circumstances. DoorDash's data handling policies for this category of content — storage, access controls, anonymization procedures — have not been made public.
On compensation fairness: the question of whether per-task payments constitute fair compensation for the value being generated is genuinely complex. A voice recording that costs DoorDash a few dollars to collect might be incorporated into a model sold as a commercial product generating millions in revenue. Workers have no visibility into or participation in that value chain beyond their initial task payout. This mirrors broader critiques of the data labeling industry, where the people who perform the foundational work of making AI systems functional are structurally excluded from the economic upside of the products they enable.
Labor economists and worker advocacy organizations have flagged these concerns in other contexts, and the Tasks launch seems likely to attract similar scrutiny as awareness of the program grows.
DoorDash is entering a market with established players, but it brings structural advantages that purpose-built data labeling platforms cannot easily replicate.
Scale AI — currently among the most prominent data annotation companies — operates a managed marketplace model where enterprise customers contract for specific dataset deliveries. Scale has raised billions in venture funding, counts major AI labs among its clients, and has built deep expertise in dataset quality assurance. Its workforce, however, is assembled from the open labor market, with all the recruitment and retention costs that entails.
Toloka, originally Yandex's internal crowdsourcing platform and now an independent entity, brings geographic diversity and a large global contributor base. Appen, despite facing revenue pressure in recent years as AI labs have sought to vertically integrate their data operations, remains a significant player in the long-tail annotation work that falls outside Scale's focus.
What none of these platforms have is a pre-existing, identity-verified, geographically distributed workforce of hundreds of thousands of people who have already passed background checks and are already engaged with the platform on a regular basis. DoorDash's marginal cost of recruiting a Tasks contributor from its existing Dasher pool is close to zero. Its marginal cost of verifying that contributor's identity and location is similarly minimal, because that infrastructure was built for food delivery.
The competitive moat here is not technology — the Tasks app itself is not technically novel. It is the distribution and onboarding infrastructure that DoorDash has spent a decade building, now being redeployed into an adjacent market.
DoorDash's Tasks app is a signal about where the gig economy is heading as much as it is a story about any particular company's product launch.
The demand for training data is not decreasing. As AI capabilities expand and applications proliferate, the need for labeled, diverse, real-world data grows in parallel. Synthetic data generation has made meaningful progress but has not eliminated the need for human-generated ground truth — particularly for the kinds of multimodal, physically grounded tasks that frontier models are now being trained to perform. Someone has to record the audio. Someone has to film the kitchen. Someone has to photograph the restaurant.
For platform companies that have spent years assembling large contractor workforces, the opportunity to monetize that workforce for data collection — especially for the periods when delivery or ride demand is low — is strategically obvious in retrospect. The surprising thing is not that DoorDash has built Tasks. The surprising thing is that it took this long.
The model DoorDash is piloting may well become standard practice across the gig economy within the next few years. Instacart workers photographing grocery shelves for retail AI. Uber drivers capturing street-level imagery for autonomous vehicle systems. TaskRabbit contractors documenting home repair procedures for robotic maintenance models. Each of these hypothetical applications follows the same template: a pre-existing, distributed human workforce, redirected toward the specific data collection needs of AI development.
The workers who power this economy deserve clarity about how their contributions are being used, fair compensation that reflects the downstream value of what they create, and legal frameworks that protect their privacy and their rights. As of March 2026, those frameworks are still catching up to the pace of deployment.
What DoorDash has built with Tasks is not just a new revenue stream or a side project. It is a preview of the labor infrastructure that will underpin the AI economy for the next decade — and a test of whether platform companies will lead with transparency or wait for regulators to force their hand.
Udit Goenka covers AI strategy, product development, and the evolving intersection of technology and labor. Follow the latest in AI at udit.co.
Cloudflare's Matthew Prince warns that automated bot and AI agent traffic will surpass human internet activity within a year, raising existential questions about the future of the web.
Amazon founder Jeff Bezos is reportedly raising $100 billion to acquire legacy manufacturing companies and transform them with AI, signaling a massive bet on industrial automation.
Senator Blackburn introduces the TRUMP AMERICA AI Act, a sweeping federal AI framework aimed at replacing fragmented state regulations with a unified national standard covering privacy, deepfakes, and child protection.