I’ve always wanted to learn to code. But I’ve never quite managed to carve out the time to do so, having been flat out as cofounder of Olio, the food sharing app. Recently I’ve had the dawning realisation that I can probably strike it off my to-do list for good. Not because I’ve done it. But because I will never need to. Coding, it seems, is on its way to becoming our generation’s calligraphy: a craft some people will continue to practise beautifully, even obsessively, but no longer the economic bottleneck around which entire organisations are designed.
This shift has profound implications for how startups are structured and how they operate. For most of the modern startup era, engineering velocity has been the limiting factor. Coding took time, and so roadmaps were constrained by developer capacity, and product ambition was filtered through what could realistically be built in a given quarter. As a founder with no shortage of product improvements in my mind’s eye, that constraint has always been deeply frustrating. But it also shaped the entire organisation: our hiring plans, business cases, culture and tempo of decision making.
With recent AI model releases, that primary constraint is well on its way to disappearing. At Olio, for example, we’ve seen the output per engineer more than double so far in the first quarter of this year vs the last quarter of 2025. That sort of step change was previously unimaginable and represents a complete paradigm shift.
We used to be able to decide what we wanted to do far faster than we could do it. Now the challenge is reversed; we need to be able to make good decisions as quickly as we can build. And that requires a very different kind of organisation.
Democratising decision-making
When build becomes dramatically faster, the scarcest engineering skill is no longer syntax, it’s judgement. Our engineers increasingly need to think like product managers, particularly on smaller, self-contained features. Instead of waiting for fully fleshed-out specs, they’re framing problems, weighing trade-offs and deciding what “good” looks like in context. That, in turn, allows our product manager to stay focused on truly cross-functional initiatives and major launches; the projects that require alignment across marketing, operations, community and partnerships.
Everyone is having to adjust their roles in real time, as we learn what these new models can really do, and we have to keep reminding ourselves that what worked last quarter may not work this quarter.
Another historic bottleneck that may finally loosen is business intelligence (BI). For years, our data access has been frustratingly constrained because we didn’t have sufficient BI capacity to meet everyone’s needs. We’ve now connected Claude directly to BigQuery, a data warehouse, and have spent some time educating it about our database structure and naming conventions. After running our BI processes in parallel during a pilot phase, we anticipate enabling the entire team to self-serve their data using natural language. This has long been the holy grail but this year it finally seems possible. And it will be totally game-changing: instead of submitting tickets or waiting for dashboards, product managers, marketers and operations leads will be able to interrogate live data themselves.
Claude’s capabilities also extend beyond product and data. Our customer support team spends significant time identifying and investigating bad actors on the platform. With AI assistance in pattern recognition, anomaly surfacing and rapid cross-referencing of behavioural signals, we can materially increase the precision and speed of those investigations. That opens up the possibility of scaling our impact without proportionally scaling customer support headcount, helping us move toward profitability faster.
While it’s phenomenal that our engineering throughput and BI capabilities are increasing so dramatically, we’re quickly learning that organisational bottlenecks don’t vanish. They migrate. In our case, the first place it has surfaced is quality assurance.
Density of complexity
We run a logic-heavy marketplace with nuanced business rules and a lot of UX branching. As build velocity has increased, so has the density of complexity. Edge cases compound faster and the surface area for unintended consequences has expanded. Historically, a team of seven engineers might have needed one QA engineer. Today, we need two. Because while AI compresses build time, it doesn’t — yet — compress complexity.
There are other bottlenecks on the horizon too: the pace of organisational decision-making and, most relevant to me, the pace of founder judgement. I’m actively thinking about how I allocate my time and shifting toward more internally focused work, because in an AI-accelerated world there’s a real risk that the organisation can execute faster than it can decide.
We will also need to upskill far more of the organisation to become generalists rather than narrow specialists, so that more of the team have the business context, and frankly the common sense, to make sound decisions without constant escalation.
Needed: clear thinkers
Play this forward and it’s clear that an AI-first world will radically change how we hire. Before opening any new role, we now ask whether AI can mitigate the need to hire at all. And if not, we hire differently. Any previous emphasis on specific “experience” or “skillsets” will be downweighted, given how dramatically the half-life of those attributes has shrunk. In their place will be a non-negotiable focus on mission alignment, values, judgement, resilience and a growth mindset.
For the past two decades, startups have always been systems optimised around their scarcest internal resource, which has been engineering capacity. Now however, the scarcity is shifting upward: toward clarity of thought, coherence of product strategy and speed of high-quality decision-making.
Coding may not disappear — just as calligraphy did not disappear — and it may even flourish as a craft. But as a structural constraint on how companies scale, its dominance is fading.
The founders who thrive in this next era therefore won’t be those who can write the cleanest code. They’ll be those who can think the clearest thoughts, and build organisations that can keep up with the machines they’ve unleashed.
