Writing code is not the bottleneck in software anymore. The harder part is deciding what is worth building, understanding the problem, making tradeoffs, and knowing what good looks like.
That raises a reasonable question about founder time. If judgment and clarity are scarce, shouldn't I spend my time defining work for the team and let other people implement it?
Definition is the bottleneck. That doesn't mean I should only define.
For a lot of meaningful product work, the process is more involved than define → implement. It looks closer to this:
idea → build → use → understand the problem better → revise → build again
I can have a clear direction and know what good looks like without having a complete specification of the finished product. Building gives me something to judge, and that judgment changes what I build next. Definition continues throughout the process.
AI makes that whole process much cheaper to run. When I have the context, the authority to decide, and the tools to execute, I can act on what I learn immediately. I can keep working through the problem while it's all still in my head, without briefing someone else or waiting for another round of implementation and review.
A workflow that passes through product management, design, development, and QA has to transfer that context at each step. As the idea changes, parts of that process repeat. Even when everyone is good at their job, those handoffs take time and attention. Making the code faster to produce doesn't remove them.
If the implementation is being done through AI in either case, I want to understand what we gain by putting those people between the idea and the result. The need for someone to write the code used to be enough to explain much of that division of work. Increasingly, it needs another reason.
What that means for founder time
If I hand off an important product problem, I may still be needed for the decisions that come up along the way. My contribution becomes a series of explanations and reviews. If I stay in the work, I can make those decisions continuously and take the idea much further in the same stretch of time. Delegation saves me some work, but it can also cost me that pace.
That matters because the value of the work isn't evenly distributed. One important product change can be worth more to the company than months of useful work elsewhere. When keeping the process together lets me pursue that opportunity much faster and further, doing it myself can be the best use of my time. The amount of work I could enable elsewhere doesn't settle the comparison.
Founders of large public companies are making time to create directly too. Shopify's Tobi Lütke said he had shipped more code in three weeks than in the previous decade with AI. Block's Jack Dorsey built Bitchat's initial prototype over a weekend using AI. Meta's Mark Zuckerberg has reportedly been spending time coding on AI projects and participating in technical reviews.
These are founders running companies with thousands of employees. If building things themselves is still worth their time, I find it hard to argue that a smaller startup's founder should focus only on managing. AI lets one person take an idea much further, with their own judgment guiding it throughout. That's what changes the return on spending time creating.
I still owe the team clear priorities and decisions. That doesn't mean building something valuable myself should automatically come second.
Where other people add value
There are good reasons to involve others. Someone may understand the customer better, find a better interaction, see an architectural problem, or catch a failure I would miss. Those are contributions that can make a handoff worthwhile.
A team also lets us pursue several important things at once. That can be worth more than moving exceptionally fast on one of them. But when the next decision on a particular problem depends on seeing the previous result, adding more people doesn't automatically make that sequence faster.
That's the distinction I want to make when thinking about scaling software development. Adding people to pursue more valuable work in parallel is different from adding people to the sequence one idea has to pass through. I want to know what each added handoff buys us.