Working with AI Means Thinking More, Not Less

The Tech Lead Metaphor

Shkuratov frames this with a striking metaphor: "I'm the tech lead, the AI is the entire team." On the surface, AI looks like a silver bullet — spinning up prototypes, handling test scaffolding, suggesting directions. The pace is extreme. But that is where the most dangerous substitution begins. A good tech lead does not stop thinking just because the team is strong; the stronger the team, the more expensive mistakes in framing and boundaries become. The responsibilities have not gone anywhere — they have become stricter.

The Real Risk: Losing Ownership of Intent

Discussions of AI in programming focus on hallucinations and bad code. But the real trouble starts when the code looks fine — clean, passing tests, exactly right. The problem is losing what Shkuratov calls "ownership of intent": the link between what you are trying to achieve, why the system is designed this way, and what distinguishes a real solution from a plausible imitation.

Once that ownership is lost, the developer faces a dangerous state: "it works, but I don't know why" — followed by "and I don't know what will break it." The failure surfaces later, during the next change, at an edge case, when reasonable local decisions collide into systemic fragility.

Why AI Forces More Disciplined Thinking

The stronger the executor, the more dangerous unclear framing becomes. The faster work gets done, the faster intent mistakes materialize. AI does not lower the demands on thinking — it makes thinking more mandatory. Developers must write goals down, separate the larger goal from the local request, define contracts for each step, read diffs, run checks, and keep in mind what stands above the code: its real-world behavior.

Anti-Drift Discipline

The human becomes a carrier of anti-drift discipline — preventing speed from turning into project drift. This means writing and rereading the goal, recording decisions in artifacts, reviewing not just results but reasoning, and checking old invariants alongside new tests. These are some of the most expensive engineering tasks — and precisely what cannot be delegated.

Seven Irreducible Responsibilities

Shkuratov distills the developer's responsibilities: holding the larger goal, breaking work into isolated parts, setting boundaries, defining signs of "done," reading everything important, running existing checks, and turning decisions into durable artifacts. The more AI can generate, the more valuable the ability to hold the task before the code exists.

Thinking, You Lead. Stop Thinking, You Get Led.

The conclusion: "Working with AI is not a mode in which you finally get to stop thinking. It is a mode in which you need to think more, not less."

More — because speed increases the price of ambiguity. More — because plausible code is more dangerous than obviously bad code. More — because someone must hold the goal, boundaries, invariants, and verification method.

"The cruel irony is that the AI knows these subtleties better than we do. But we still have to ask. We still have to frame the question. We still have to notice that a question is needed at all. And we still have to do the thinking."

AI is not permission to switch off your brain. It is an invitation to bring your best thinking forward — because the cost of not doing so has never been higher.