Part 1 of 3 in our Vercel Ship 2026 series. By Nathan Connolly, CEO of Consortia.
On 17 June I spent the day at Vercel Ship in London. It is one of the larger event that brings together many of the people building the next generation of software products across the UK.
I came away from the day with a notebook full of product detail, but the my biggest takeaways weren't neccessarily technical and sit much closer to my day job. It was what these changes mean for the people companies need to hire.
The way companies build is changing rapidly, and that changes the people they need. Here is what stood out, and what I think it means for anyone hiring across Product, Data and AI, Engineering, and UX.
Software now has to be built for agents, not just people
One of the opening topics was around agents becoming the primary product users. For years teams have designed tools, documents, and interfaces for a person to read and click. But the discussion at Vercel Ship focused on a different reality where software also needs to work for AI systems acting on a user's behalf.
One retailer described creating structured content specifically for AI systems to discover and interpret. The goal was not just to improve the human experience, but to make information easier for AI agents to discover and act upon. While it sounds like a small change in practice it has wider implications because when the end user might be a machine, product design, data architecture and content strategy all start to look very different.
For us and for hiring, that's where it gets interesting. More companies are looking for people who understand both audiences, needing individuals who can create experiences for humans while also understanding how agents consume information and complete tasks.
Many organisations are already looking for those skills, even if they haven't yet updated the job description to reflect it.
The "harness" matters more than the model
The most repeated line of the day was “the model is not the hard part anymore”. The hard part is everything around it. How you manage context, connect tools, handle errors, and keep an agent on track. The industry increasingly refers to this surrounding infrastructure as the "harness", everything that sits around the model and determines whether it produces reliable outcomes.
The team from Cursor made the point well. They explained how their own models now compete on quality with some of the strongest alternatives available, often at a significantly lower cost. Their advantage comes less from the model itself and more from how they deploy and manage it.
Companies that understand this are hiring for a different profile than the one they advertised even a year ago. The people who stand out tend to combine technical capability with the ability to think about process, governance and business outcomes.
Autonomy is a question of risk and reversibility
This did lead to a sensible question raised though: How much should you actually let an agent do on its own? The answer that seemed to come up repeatedly was to govern autonomy by risk level and ability to reverse an event.
A large, well-tested change might be safe to automate end-to-end. A two-line change to a login system is not. Drafting an email is easily reversible, but moving money is not. Several speakers described sandboxing an agent first and gradually giving it more access as trust is earned. All the while, making sure there is a clear record of what the agent did if something does go wrong. One company shared that 60 to 70% of their code changes now merge automatically, with humans focused on the riskier minority.
It made me think about what this means for leadership and hiring. Good leaders have always had to decide what to delegate, what needs oversight and where the biggest risks sit. AI simply introduces another layer to those decisions. The technology may be different, but the mindset needed for leaders isn't.
The human role doesn't disappear
That naturally led to another question that came up throughout the day. If agents are becoming more capable, what is left for people to do other than delegate?
Models are very good at finding patterns in what they already know. They're much less reliable when the task requires questioning those assumptions or recognising that the available information may itself be incomplete. That's where people still make the difference.
During an AI panel I spoke on last year, I made a similar point. We can't just accept the AI output because it sounds convincing. We need to question the answer and the source abd find out why the model reached that conclusion in the first place.
As AI becomes part of everyday work, the people who create the most value won't be the ones who simply use it well. They'll be the ones who know when to push back, test assumptions and spot the gaps that a model can't see.
For me, that's one of the biggest misconceptions around AI. The goal isn't to stop thinking because the technology can generate an answer. It's to ask better questions and recognise when a different answer is needed.
The proof is already in production
After so much discussion about trust, risk and where AI should be given autonomy, what struck me was how few conversations were actually about the future. Most of the examples being shared were already live. AI agents weren't being talked about as experiments or prototypes. They were already part of day-to-day operations. Two examples in particular stuck with me.
How a two-person team built an in-house support agent in three weeks that now resolves 91% of customer tickets on its own, against an industry benchmark of 60 to 75%.
That an autonomous sales development agent trained on the behaviour of its best-performing rep had generated a reported 32x return and is maintained by a single engineer.
When small teams can deliver at that scale, every hire carries more weight. That's something we're already seeing in the briefs coming into Consortia. Businesses are thinking less about headcount and more about finding people who can combine technical expertise with sound judgement and make the most of the tools now available to them.
Closing Thoughts
Walking away from the event, I kept coming back to one thought. While AI is moving incredibly quickly, the value of good judgement, curiosity and human insight isn't shrinking. If anything, it's clear it's becoming more important.
Across Product, Data and AI, Engineering, and UX, the briefs we see have been shifting from "can you do the task" to "can you decide what is worth doing, and can you tell when the machine has it wrong." These are much harder things to screen for, as they aren’t technical.
Technical capability still matters, but employers are placing more value on systems thinking, commercial awareness and sound judgement alongside technical expertise.
Those aren't qualities that are easy to spot from a CV alone, but they're increasingly what separates exceptional hires from simply capable ones.
In Part 2, I'll look at the story behind that 91% support agent and what it says about the way high-leverage engineering teams are being built. Then, in Part 3, I'll explore autonomy, trust and why governance is quickly becoming as much a hiring question as a technical one.
These conversations are moving quickly, and they're already starting to influence the hiring decisions we see every day. If you're thinking about what this means for your own team, we'd be happy to share what we're seeing in the market across Product, Data & AI, Engineering and UX, and where demand is beginning to shift.