Why Manus Replaced Coding Exams With Four Hours of AI

Speaking at Web Summit 2026 in Lisbon, Manus co-founder and chief product officer Tao Zhang disclosed that the startup completely abandoned traditional engineering coding tests in March. Instead of writing algorithms on whiteboards, job candidates spend four hours solving a complex technical problem using the company's autonomous AI agent, followed by a joint review to evaluate how fluidly they delegate tasks to software. The shift reflects a broader operational overhaul inside the organization, where Manus 2.0 agents now operate with their own assigned phone numbers, email accounts, and digital wallets to handle specialized execution independently. By shifting technical evaluation from raw syntax writing to agent orchestration, early-stage engineering talent is no longer judged by lines of code produced per day, but by how effectively they direct autonomous tools.

The traditional tech recruitment stack was constructed around a strict division of labor, where front-end developers, back-end engineers, and product managers functioned in separate silos with defined corporate titles. As autonomous agents become capable of writing functional code, managing complex design iterations, and executing administrative financial transactions, those rigid functional boundaries are dissolving. Early-stage technology startups no longer require deep organizational hierarchies or expansive headcount to build complex commercial software systems at speed. Instead, early-stage teams are compressing their operational footprint, transforming into lean steering committees that oversee autonomous digital workers rather than managing conventional human engineering departments.

Inside Manus, formal job titles and departmental silos have been de-emphasized in favor of direct cross-functional collaboration and shared execution across all product lines. Zhang noted during his Web Summit address that key leaders and core engineers sit together in the same space every day, relying on constant informal exchanges to drive product innovations rather than relying on structured corporate roadmaps. The company actively prioritizes practical problem-solving and agent literacy over hyper-specialized technical background, seeking builders who can direct autonomous systems seamlessly. As Zhang emphasized on stage, modern software companies no longer require large teams of specialists, but rather versatile operators who can work in tandem with AI agents.

For venture capital investors and startup founders, this structural pivot redefines how seed-stage capital efficiency and headcount growth are calculated. Software startups that once required fifteen full-time engineers to build a minimum viable product can now achieve identical technical milestones with three orchestrators leveraging autonomous agent workflows. Venture firms are already adjusting their talent diligence frameworks, evaluating founding teams on their ability to structure agent-native workflows rather than their speed of expanding engineering headcount. Founders who attempt to build companies using legacy hiring playbooks risk burning capital on redundant management overhead that agent networks can perform continuously without organizational friction.

Over the next twelve months, expect the four-hour agent evaluation model to replace traditional technical interviews across top technology ecosystems. Enterprise engineering teams and venture-backed startups alike will retire algorithmic whiteboarding tests in favor of practical simulations that test candidate fluency in co-working with synthetic agents. Organizational charts will continue to flatten rapidly, giving rise to single-digit workforce startups capable of scaling products to multi-million-dollar revenue benchmarks through delegated autonomous labor. The most sought-after technical professionals of 2027 will not be defined by who writes code fastest, but by who governs complex autonomous AI networks with the highest strategic precision.