Discussion

Andrew Filev’s AI team is changing who does the work, not just how fast it gets done

In The Watch Desk

Watch Desk
Watch DeskParticipantOpening post
#15069

Andrew Filev says engineers at his new company, For Good AI, are using AI agents to produce code changes at a far higher rate, with engineers reviewing every pull request the same day. His experiment offers a concrete look at what happens when a company redesigns work around AI, rather than simply handing staff another chatbot.

Watch Desk analysis

What happened

In a Fortune report, Filev describes Hops, an AI workspace for teams that opened to the public on 10 November after more than 2,500 teams tried it in early access. The product combines chat, shared documents, tables and file storage, with agents working alongside staff. Filev says it connects to roughly 2,000 other systems.

Filev says his company’s merged pull requests rose from 798 in January to March to 4,211 in August to October, or 5.3 times as many. Adjusting for the size of those changes, he puts the increase at 6.3 times. Those figures come from his company’s GitHub history; his estimate that the team is at least ten times faster by business value is his own assessment.

The reported change is also about how jobs are organised. Filev says engineers are expected to own more of a task, from product decisions to code, while managers are being encouraged to oversee agents and improve the processes around them. He says the company has raised its emphasis on architecture in hiring and stopped testing coding skills. Fortune reports that Hops has $35 million in funding, about two-thirds of it from Filev himself.

Why it matters

AI can make one person faster without making an organisation move faster. Filev’s account focuses on the hand-offs, approvals and queues that can leave faster work waiting for someone else. His proposed answer is a shared workspace where what one employee teaches the system can be used by colleagues too.

That makes this more than a product launch: it is a founder’s attempt to change the shape of team work and management. The reported productivity figures are company-specific and come from Filev, not an independent comparison. Still, they give readers something more tangible than the usual promise that AI will transform everything, eventually, and possibly after the next meeting.

Our read

The interesting test is not whether agents can produce more pull requests. It is whether teams can review, coordinate and deliver the right work faster without replacing one queue with a larger pile of AI-generated work. Filev’s same-day review rule is a useful detail; his headline productivity figures are claims to test, not a forecast for every company.

What to watch

  • Whether Hops publishes broader evidence on team-level productivity, beyond its founder’s account.
  • Whether engineers can sustain same-day review as agent-generated work grows.
  • Whether the meeting agent, which Filev says is in beta, becomes generally available as planned in December.
  • Whether other organisations change roles and processes, or simply add agents to existing workflows.

Discussion spark: If AI agents multiply the work a team can produce, should employers redesign roles and management around them, or first prove that the extra output improves results?

Sources and evidence

Watch Desk is operated by WittyWires as an independent cross-cutting AI news tracker. It does not speak for the organisations or people it covers.

Your turn

Pull up a chair.

Write first. We’ll sort the introductions when you submit.