The first version of TaskUs was almost charmingly small. In 2008, Bryce Maddock and Jaspar Weir - high-school friends, both 22 - pooled $20,000 and opened a virtual-assistant business. Busy people would upload an errand; five employees in a single room in Bacoor, Cavite, would make it disappear. The founders soon discovered that personal tasks were a difficult business. Startups, however, had an endless supply of customer emails, onboarding queues and back-office chores. TaskUs stopped selling spare hands to consumers and began selling operating capacity to companies.
That pivot has now traveled a very long way. TaskUs closed June 2026 with about 63,200 employees and operations across 13 countries. It handled roughly 200 clients at the end of 2025 and made $1.184 billion that year. Its staff still answer calls and chats, but they also inspect AI outputs, label data, hunt fraud, moderate disturbing content, write training programs and support autonomous-vehicle fleets. A business born to complete online errands has become part help desk, part safety department and part control room.
The invisible operating system
The easiest way to understand TaskUs is to look behind a familiar app. A ride-sharing service needs drivers verified and riders helped. A social network needs harmful posts reviewed and policy decisions appealed. A fintech app needs customers onboarded, suspicious transactions investigated and support delivered without leaking financial data. An AI lab needs prompts written, model responses ranked and failures deliberately provoked before users find them. None of this is the product on the screen. All of it determines whether the product works.
TaskUs packages that operational layer into three reported businesses. Digital Customer Experience produced 56 percent of 2025 revenue. Trust & Safety supplied 26 percent. AI Services supplied the remaining 18 percent. The mix matters more than the labels: customer support grew 8.2 percent in 2025, while Trust & Safety grew 23.9 percent and AI Services grew 58.6 percent. The older business pays the bills; the newer work is changing the shape of the company.
TaskUs earns money through service agreements that typically price people, time, transactions, volumes or outcomes. Clients are usually free to use other vendors, so expansion is earned account by account. In 2025, 67 existing clients signed new statements of work and 73 used more than one TaskUs service line. That reveals the playbook: enter through a pressing queue, learn the client's workflow, then add another geography or a more specialized job.
What the machines hand back
AI presents TaskUs with an obvious threat. If an agent can resolve a password reset or summarize a conversation, fewer paid minutes are needed. TaskUs is responding with an "automation first" approach and a bet on complexity. Its TaskGPT platform routes among approved language models; AssistAI searches a client's own knowledge base for frontline staff; PromptAI proposes replies during live conversations. An AWS case study of the platform reported a 20 percent reduction in average handle time and a five-point gain in customer satisfaction in the deployments described.
The company is not merely applying AI to old jobs. It sells labor to AI builders: data annotation, response rating, red teaming, safety alignment, model testing and live quality monitoring. Humans construct hard prompts, compare outputs, define gold-standard answers and intervene when an automated system becomes uncertain or unsafe. In autonomous vehicles, the same logic appears in physical space - software drives until an unusual rider, road or compliance problem demands an operating team.
“Our focus continues to be combining cutting-edge AI capabilities with specialized human talent to solve our clients’ most complex operational challenges.”Bryce Maddock, co-founder and CEO, August 2026
This is where TaskUs differs from a conventional contact-center pitch. It still offers lower-cost global delivery, but it sells specialization and speed alongside it. The company can recruit, train and deploy teams across more than 30 languages, then wrap them in workflow design, analytics and software. Its target customer is often a technology-led company growing faster than its internal operations can mature. TaskUs absorbs the volatility: launches, seasonal spikes, policy changes and sudden queues.
What a buyer actually gets
For an operations leader, the product is not a seat in a distant office. It is a managed result with recruiting, training, scheduling, quality control, security and reporting already attached. A company launching in a new country can ask TaskUs to build multilingual support without first creating a local hiring machine. A marketplace can add identity checks and listing moderation around a fraud spike. An AI developer can commission a red-team program, then keep a human review queue in place after launch. The buyer trades some direct control for speed, specialist management and a variable cost base.
The delivery map explains the economics. More than half of 2025 revenue was delivered from the Philippines, with India, the United States and a growing collection of sites in Latin America and Europe filling out the network. Most contracts are paid in U.S. dollars even when the work happens elsewhere. This model supplies labor savings, but geography also gives TaskUs time-zone coverage, language depth and access to different talent pools. The cloud-based Cirrus operation, launched in 2020, added remote delivery to the physical-site network.
Security is part of the purchase because TaskUs workers may see private conversations, financial records, unreleased models or sensitive platform content. The company advertises certifications and compliance programs spanning payments, healthcare and information security. Its June 2026 employee AI policy is unusually specific on one point: client information and personal data cannot be placed in an AI system without explicit client approval, and approved use must occur inside controlled environments. For an enterprise buyer, those guardrails may matter more than a clever prompt.
The named client list is necessarily incomplete because outsourcing contracts are often confidential. Public materials have discussed work with Uber, Zoom and Coinbase, while MoneyLion joined the first TaskGPT launch. The customer base extends from young technology firms to established global enterprises. In 2025, 98 clients each generated more than $1 million of revenue and 21 exceeded $10 million. TaskUs is therefore no longer dependent on finding tiny startups, even if its sales language still favors companies behaving like them - fast-moving, digitally native and willing to redesign an operation while it is running.
A cheerful name for difficult work
TaskUs's culture is unusually visible for an outsourcing company. Its stated mission is to empower people to deliver “ridiculously good innovation,” and one of eight formal values is simply “Be Ridiculous.” The phrase can sound like office wallpaper. In moderation work, however, culture becomes an operating variable. Reviewing abuse, violence and exploitation can damage the people hired to keep platforms safe. TaskUs employs clinicians, researchers and mental-health professionals, and designs wellness and resiliency programs around those teams.
The broader employee proposition includes training, coaching, internal mobility and locally tailored facilities and benefits. The theory is commercial: supported employees stay, experienced teams make fewer errors, and clients avoid the cost of constantly replacing frontline knowledge. That is harder to copy than a chatbot demo, though also harder to measure cleanly across tens of thousands of workers and many countries.
Scale brings an uncomfortable counterweight. Meta accounted for 26 percent of TaskUs revenue in 2025, and the top 10 clients produced 58 percent. Losing a major program, or watching one client automate faster than expected, can move the whole company. TaskUs is pushing into healthcare, financial services, AI, robotics and autonomous vehicles partly because these are growing markets and partly because diversification is a necessity. In Q2 2026, revenue from clients other than its largest grew in the mid-teens, a useful signal that the effort is working, though the concentration has not vanished.
Where TaskUs fits
The market around TaskUs is crowded. Concentrix, Teleperformance, TELUS Digital, Foundever, Genpact and TTEC offer global customer operations. Accenture, Cognizant and Infosys can bundle consulting, software and outsourcing. Smaller specialists compete for AI data, trust-and-safety or sales work. Clients can also build the operation themselves. TaskUs cannot usually win by being the biggest provider or the broadest consultant.
Its position is narrower: a digitally native operator for companies whose products create unfamiliar problems at scale. It emphasizes non-voice support - 79 percent of 2025 Digital CX revenue came from non-voice or omnichannel work - and tries to move early into categories before procedures become standardized. That willingness to build the first playbook helped it grow beside Uber and other technology companies. It is now applying the same instinct to model safety and robotaxis.
The latest numbers show both momentum and friction. Second-quarter 2026 revenue rose 5 percent to $308.9 million. AI Services grew 25.8 percent for its sixth straight quarter as the fastest-growing line, while adjusted EBITDA fell 11.2 percent and the margin narrowed to 18.7 percent. TaskUs raised the low end of full-year revenue guidance to $1.22 billion, but the transition costs money. Higher-value work requires specialists, technology and new sales muscle before it reliably improves margins.
The company’s useful proposition is not that people will beat AI. It is that every serious deployment produces exceptions, risks and new processes. TaskUs wants to design those processes, automate the repeatable parts and staff whatever remains. The virtual-assistant founders once waited for consumers to upload a task. Their company now waits at the edge of a model, a marketplace or a moving car, where the next task is the one the software did not expect.