Conclusion: the craft of confidence
Estimated read time: 6 minutes
In the end, this paper isn’t really about AI. It’s about confidence.
Not hype-fuelled bravado. Not the brittle confidence that comes from pretending we already know the answer. The kind of confidence public service actually needs: earned through evidence, inclusion, and diligent practice – the confidence to move faster because we trust the methods we’ve honed over many years, methods that keep us honest and our users at the centre.
In a vibe-coded world, building gets cheaper. Confidence doesn’t. We can make things real faster than before, but the cost of being wrong hasn’t fallen.
A vibe-coded world compresses the distance between intent and reality. That creates an opportunity, and a temptation. The opportunity is that we can deliver value earlier, learn sooner, and stop wasting months defending guesses. The temptation is that speed becomes a substitute for thought, or that we treat “working software” as if it were the same thing as a legitimate, supportable service. The craft is not letting speed blur that line: using it to get closer to reality, not further from it.
A week that’s now plausible
Picture a week that would have sounded implausible in government not long ago, but now sits within reach.
A team notices a spike in confusion and avoidable contact. By the afternoon they’ve designed and shipped a small change to a safe environment. The next day it’s in front of users and frontline colleagues and they observe what happens. Then it’s out in the open, behind a feature flag for a small percentage of live traffic, actively monitored, with rollback ready if needed. Twenty-four hours later the evidence is clear about whether the change helps or doesn’t; and if it doesn’t, it’s removed without drama, with the learning captured. By the end of the week the team can reflect: this is what we tried, this is what changed, this is what we learned, and this is what we will do next.
Nothing in that rhythm is reckless. It is cautious where it should be, and bold where it can be. The speed comes from smaller steps, clearer evidence, and less waiting – not from gambling, not from heroics, and not from “trust us”.
That is what I mean by the craft of confidence: designing a way of working where each step is small enough to be safe, visible enough to be accountable, and evidence-rich enough to justify the next move. It’s how you “seek forgiveness not permission” without drifting into arrogance, because you are not bypassing governance; you are meeting it continuously, in public, through what you ship and what you measure.
Confidence for all
This is also why this isn’t only a product or engineering story. In a vibe-coded world, more of the multidisciplinary team can express intent as runnable artefacts. Confidence has to be built in teams, yes, but also in the people who fund work, assure it, operate it, support it, and ultimately answer for it.
That has implications.
For leaders, the question shifts. Not “did you deliver to the plan?” but “did you earn the right to go faster?” Did you build a system that can learn safely at pace? Did you make the safe path the easiest path? Did you invest in the on-ramps (platforms, environments, governance built into delivery) so teams aren’t forced to choose between speed and legitimacy?
For practitioners, it’s similar. The best response to scepticism or anxiety is not a manifesto; it is practice. Try the tools. Use them on low-risk work. Notice where they help and where they mislead. Then double-down on what they cannot do: empathise, exercise judgement, weigh trade-offs, notice who is excluded, and defend what “good” means. If AI compresses the time to build, then human responsibility becomes the scarce resource we protect.
And for the public, ideally, most of this is invisible. People do not need to know what tools we used. They need to experience services that work: clear, accessible, reliable, humane; improving steadily; honest about mistakes; quick to fix what breaks. If we do this well, trust can grow not because we promise less change, but because we demonstrate competence in change.
Humility as a discipline
There is one more dimension to confidence that matters here: humility.
Real confidence is not certainty. It is the ability to act without pretending to be infallible. An AI-assisted world will make our wrongness show up faster, which is a gift if we are willing to see it. The iterative discipline has always implied “we might be wrong, so we test”. AI-assisted delivery makes that easier than ever, because testing is cheaper and learning loops are tighter. But that only helps if we keep our curiosity alive: curiosity about what users actually experience, curiosity about what the data is really saying, curiosity about what we don’t yet understand, and the willingness to slow down when the evidence demands it.
If you want a summary of what this paper is arguing, it is this: a vibe-coded world doesn’t invent new values; it extends our capacity to live up to our disciplines. It reduces the distance between policy intent and experienced reality.
If you want something to take away, I think this paper boils down to this:
- Have a go, but keep it bounded and honest. Use AI on low-risk problems first. Time-box experiments, name the hypothesis, capture what happened, and discard without shame.
- Do not mistake “it runs” for “it’s ready”. Working code is cheaper now. Readiness is still expensive: accessibility, security, lawful data handling, operability, monitoring, support paths, and rollback. Protect teams when they say “not yet”.
- Make proof the currency of progress. Replace confidence-by-document with evidence-by-default: show the thing, measure the effect, record what you learned, and let that trail earn autonomy, funding, and the right to scale.
- Put standards and governance in the machinery, not in ceremonies. If more people can ship, guardrails must live in defaults: pipelines, platforms, component libraries, feature flags, and transparent decision logs, with policy, legal and ops close to the team, not waiting at the end.
- Let outcomes set the pace, and be willing to undo your own work. Organise around the numbers that matter, be explicit about uncertainty, and scale only when the evidence holds. When it doesn’t, stop early, and publish what you learned.
That is how you go faster without becoming reckless, and make it easier for the next team: less faith in plans, more confidence in proof.
The future of public sector product management in a vibe-coded world is not AI doing everything. It is human values, human judgement, and human collaboration mattering even more, at a higher cadence, with less waiting to hide sloppy thinking, and fewer excuses not to learn in public.
That is a demanding future. But it is also a hopeful one: a future where we can improve services faster than we once thought possible without losing our integrity, because we have learned how to build confidence as a craft.