AI Agents: The New Corporate Middle Managers
At a London AI conference, it became clear: software engineers are now managing AI agents. The future? Less coding, more oversight.
The AI conference in London was a revelation. Engineers from Google, Anthropic, and OpenAI gathered to discuss one thing: agents. And it seems we're all becoming managers of these digital helpers.
Gone are the days when engineers just wrote code. Now, they're steering and correcting AI agents, making sure they don't go rogue. Ryan Lopopolo from OpenAI nailed it when he said that coding's changed massively since late 2025. Today, engineers are more about unblocking and guiding agents than crafting lines of code.
The Shift in Engineering
This shift raises key questions: How much control do we really want to give these agents? Are our human languages even enough to tell them what to do? And should they've their own little helper agents too?
David Soria Parra from Anthropic predicts that 2026 will be the year we put these agents into full production. It's not just a coder's problem anymore. We're talking about agents taking over tasks like financial analysis and marketing. Imagine that. Less grunt work for us, but with a catch: they need oversight.
Managing the Machines
Funny thing is, while companies like Meta, Google, and Amazon are cutting management layers, we might end up with everyone acting as AI supervisors. Individual contributors, once free from the burden of managing people, now check and delegate work done by AI.
Mario Zechner, who created the coding agent Pi, was a voice of caution. He warned against letting agents decide too much since their decisions are based on internet-learned data, which isn't always trustworthy. His advice? Use them sparingly and don't fully trust them to make decisions for you.
Agentcraft and the Future
What should these agents even look like? Ido Salomon from Monday.com gave us a glimpse with his creation 'Agentcraft', imagine managing agents in a 'Warcraft'-styled environment. It’s a bit surreal, but maybe that's where we're headed. With heatmaps showing if agents are at risk of clashing, it looks like we’re turning agent management into a video game.
The reality is, managing these AI agents might even mean redesigning parts of the web. Vercel's CTO, Malte Ubl, pointed out that a whopping 60% of page views on their site came from agents. With software increasingly being used by agents, the big question is: how will they interact with our digital world?
Get AI news in your inbox
Daily digest of what matters in AI.