It's been roughly a year since I first asked an AI if it could read my screen. In that time, I've shipped three SaaS products, rebuilt legacy systems, designed product UXes through conversation, automated bug triage, and fundamentally changed how I write software. It felt like a good moment to step back and ask a question that's been nagging at me: who's actually making the decisions?
Because from the outside, it might look like the AI is driving. The code generation is AI. The refactoring is AI. The test writing, the deployment configuration, the component scaffolding - increasingly AI. If you watched a screen recording of my typical workday, you'd see Claude Code producing code at a rate no human could match. It would be easy to conclude that the human is becoming redundant.
After a year of doing this, I can tell you with confidence: that conclusion is wrong. And not in a wishful-thinking, "humans will always be special" way. In a specific, demonstrable, observable way.
The Creative Breakthroughs Were All Human
I went back through my projects and tracked where the ideas came from. Not the implementation ideas - not "use a repository pattern" or "add a loading spinner here" - but the product ideas. The features that differentiated a product. The design decisions that made something feel right instead of just functional.
Every single one came from me.
The Inbox concept for TestPlan - the idea that testers should have a centralised place showing everything that needs their attention. That was mine. I described it to Claude in a sentence, and Claude built it brilliantly. But the concept, the product insight that testers don't want to hunt for work? That came from years of watching people use QA tools and noticing what frustrated them.
The decision to remove the payment section from a complex multi-use interface and replace it with a dedicated modal flow. That conversation started with me saying "hear me out... how about we remove the payment bit entirely from this screen and use a modal flow instead?" The AI implemented it flawlessly. But the insight - that the interface was trying to do too many things at once and needed to be decomposed - came from product thinking, not technical thinking.
The persona framework for UX audits - creating detailed user personas and then systematically evaluating every screen from each persona's perspective. That idea was mine. Claude executed the audits with impressive thoroughness, but the methodology was human-invented.
Pricing strategy, product positioning, when to pivot, what to build next, which features to cut - all human decisions.
Why This Matters
I'll be honest with you - I find this reassuring. Not because I need to feel important, but because it answers a question that I think a lot of developers are quietly worrying about: if AI can write the code, what exactly is my role?
Your role is the same as it's always been, just with different tools. You're the product manager. You're the one who understands the users, the market, the business context. You're the one who has the creative insights that turn a technically competent application into something people actually want to use.
AI doesn't have opinions about whether a feature is a good idea. It'll build whatever you ask for, with equal enthusiasm and competence, whether it's brilliant or pointless. It won't tell you that your users don't need another settings page. It won't notice that your onboarding flow has too many steps. It won't feel the difference between a product that's merely functional and one that's genuinely good.
That feeling - that product sense - is still entirely human territory.
The Moments Where AI Pushed Back
I should qualify the above, because there were genuinely useful moments where Claude challenged my thinking. Not product vision challenges - strategic thinking challenges.
The most memorable was during CoSurf's development, when I was getting excited about features and Claude said: "You're falling into a classic trap. You're adding features to a product that doesn't yet have a clear answer to the fundamental question: who is this for?"
That's not a product idea. It's a strategic observation. And it was exactly right. I was feature-creeping before I'd nailed the core value proposition. The AI didn't tell me what the value proposition should be - that remained my job - but it identified that I was skipping a step.
There was another moment when I asked about prioritisation across all three products, and Claude came back with: "Honest question - are you spreading yourself across three products when focusing on one would get you to meaningful revenue faster?" Ouch. Also fair. I didn't take the advice, because I'm constitutionally incapable of focusing on one thing, but the question was valuable.
So the AI isn't purely passive. It's a strategic sounding board. It can identify patterns you're too close to see, challenge assumptions you haven't examined, and ask uncomfortable questions. But it's doing this as a thinking partner, not as a decision maker. The decisions remain human.
What AI Is Actually Good At
If AI isn't the product manager, what is it? After a year, I'd describe it as the most capable development team a solo founder has ever had access to.
It's the architect who can evaluate six different approaches to a problem and articulate the trade-offs of each. It's the full-stack developer who can implement a feature from database schema to API to UI component in a fraction of the time a human would take. It's the code reviewer who catches patterns and anti-patterns. It's the documentation writer who never complains about boring work. It's the QA engineer who can generate test cases systematically.
What it is not, and what I don't think it will be for a long time, is the person who walks into a room and says "I think we should build an inbox for testers." That leap - from observing a problem to imagining a solution that doesn't exist yet - is still a human capability.
The PM/Developer Dynamic
The most productive pattern I've found is treating the human-AI relationship as a product manager / development team dynamic. And I mean that literally, not metaphorically.
I write the requirements. Not formal requirements documents - conversational descriptions of what I want and why. I make the scope decisions. I prioritise. I say "no, not that, this" when the implementation doesn't match my vision. I decide when something is done.
The AI builds. It proposes technical approaches. It raises technical concerns. It implements, tests, and iterates. It's remarkably good at all of this.
The dynamic works because both sides are doing what they're best at. I'm not wasting time on implementation details that the AI handles better and faster than I could. The AI isn't wasting tokens trying to be creative about product direction when that's not its strength.
If anything, AI has made the product management skills more valuable, not less. When implementation is cheap and fast, the bottleneck shifts to knowing what to implement. The developer who can think like a product manager - who understands users, who has taste, who can make judgment calls about what matters - is more valuable in an AI-assisted world than in a manual-coding world.
A Year of Evidence
I've built three SaaS products. I've rebuilt client systems. I've designed dozens of features. I've made hundreds of product decisions. And looking back at all of it, the pattern is consistent:
Every creative breakthrough was human. The Inbox. The modal flow redesign. The persona framework. The onboarding tour concept. The analytics dashboard design. The pricing models. All human.
Every execution breakthrough was AI. The speed of implementation. The breadth of technical knowledge. The ability to refactor entire codebases. The systematic test generation. The tireless iteration on UI polish. All AI.
The best outcomes happened when both were working together. Human vision, AI execution. Human taste, AI speed. Human judgment, AI capability.
What This Means Going Forward
I think the anxiety about AI replacing developers is misplaced. Not wrong exactly - some kinds of development work will absolutely be automated away. But the developers who have product sense, who understand users, who can make creative leaps - those developers aren't being replaced. They're being amplified.
The role is shifting. Less time typing code, more time thinking about what the code should do. Less time in the IDE, more time with users. Less time on implementation, more time on product strategy. These are not lesser skills. Arguably, they're the skills that always mattered most - we just couldn't afford to focus on them because the implementation consumed all our time.
After a year of AI-first development, the human is still the product manager. The human is still the one with the vision, the taste, and the judgment. The AI is the most extraordinary development partner I've ever worked with.
But it's a partner, not a replacement. And I think that's exactly how it should be.
