60% of Tech Team Work Will Belong to AI Agents, Artefact CTO Says

A senior AI executive has put a number on the shift he says every people leader must plan for: in a mature tech team, 60% of work will be done by agents, 25% by augmented humans and 15% by humans alone. Victor Coimbra, CTO and Managing Partner at data and AI consultancy Artefact, offered that split in an interview ahead of the firm's Adopt AI Summit, scheduled for December 3–4, 2026, at the Grand Palais in Paris.

His core message for employers is blunt. Many AI pilots stall because organizations bolt the technology onto structures designed for humans. "The technology usually works," Coimbra said. "What is often still open is the division of work: which tasks belong to agents, which stay with humans, who reviews, and who owns the result."

Why do AI pilots fail to reach production?

Coimbra identified three needs that slow scale-up: token costs that explode under usage-based pricing, security and governance gaps, and adoption that never converts access into real usage. Executives buy a change in a P&L line, he learned at Bain and Accenture, yet much of that value gets lost between the slide deck and production.

Companies that scale treat AI as organizational redesign, not a tool purchase. Coimbra described two loops. In the inner loop, agents execute end to end: code, build, test, push, review. In the outer loop, humans set intent, verify results and monitor cost and quality. One shared platform sits underneath and serves every use case.

How should teams of humans and agents be structured?

Coimbra said the hybrid organization already exists, and dividing the workforce between humans and agents is the key to scaling. In his model, the org chart allocates cognitive capacity, not headcount. A Head of IT sets intent and owns outcome and cost. A senior developer verifies agent deliveries and decides architecture. An orchestrator agent routes tasks, and a review agent checks every delivery.

The productivity gains compound in three steps, he said:

• Augmented employees deliver around 5x output.
• Agents working end to end as synthetic employees deliver around 10x.
• Orchestrating both as one workforce is where the 1000x potential lies.

Training follows the same logic. People learn to brief an agent the way they would brief a junior engineer, with structured context and clear outcomes, then spend their own time on taste, planning and decisions.

What work belongs to agents, and what stays human?

Coimbra drew the line by direction of sight. Agents excel at work that looks to the past: analytics, coding, testing, documentation, reconciliation. Humans own work that looks to the future: vision, planning, deciding and judging whether a result is good. Safeguards must be engineered, not prompted. Artefact uses six harness layers, including a sandboxed runtime, checkpoints, model routing and guardrails such as human approvals, policy checks and PII controls. Autonomy grows with track record, the same way it does for a new hire.

Context is the layer employers should own completely, he argued. Agents without it follow written rules and miss unwritten ones, retrieve wrong information with total confidence, or work from last quarter's facts. He advised capturing knowledge from the work itself — meetings, tickets, decisions, corrections — rather than from documentation projects that age quickly.

How should HR and IT measure the return?

Coimbra urged leaders to measure the hybrid workforce, not the tool, against a baseline taken before rollout. His four levels:

• Adoption: do people use it?
• Productivity: lead time, throughput, resolution time.
• Quality: escaped defects, change failure rate, reework.
• Value and cost: effort freed, time to market, full cost per task.

One overlooked metric, he said, is how cognitive capacity moves between humans and agents, team by team. Freed human capacity must be redeployed to future-facing work with intent, or part of the gain is lost. He also recommended treating AI spend like cloud spend — visible, attributed to a team, tied to value — with FinOps built into the platform from day one.

Sector differences set the pace. In retail, mistakes are cheap and agents take work faster. In financial services and healthcare, every decision must be explainable, so humans keep verification longer. Compliance works best as a property of the platform, with tiered oversight that expands as agents earn trust.

Looking ahead, Coimbra predicted companies will move from counting headcount to managing shared cognitive capacity between humans and agents. "The companies that solve the division of work first will lead the way," he said — the conversation he intends to anchor in Paris this December.