In a related piece, we made the case that the modern software agency isn’t AI or human developers, it’s both, working as a single, integrated unit. That’s the theory. This is what it looks like in practice.
Consider a pattern that plays out constantly in modern development: a production bug on a live client site. It’s a useful example precisely because it’s ordinary. The kind of problem that, handled the old way, eats an afternoon. Watch where AI provides leverage and where the human stays firmly in control.
1. TRIAGE AND RESEARCH
A contact form stops submitting on a password-protected page. The developer doesn’t start guessing. They describe the symptom to the AI (form silently fails, page is protected, no console errors) and ask what conditions are known to cause it.
In seconds, the AI surfaces several candidates, including how protected-page POST requests can be stripped of their session context. The developer recognizes this as the likely culprit. This is a judgment call the AI can’t make on its own, because only the developer knows how this particular page is actually configured.
2. DIAGNOSIS
The developer feeds the AI the relevant template and form configuration and asks it to trace the request path. The AI maps the flow and confirms the request is being intercepted before the handler fires.
Now the human sets the constraint: fix this without disabling password protection, and don’t touch the global form settings. This is guidance the AI would never supply on its own. It doesn’t know which trade-offs are acceptable to the business.
3. IMPLEMENTATION
Working inside that constraint, the AI proposes switching the form to AJAX submission and supplies the exact change. The developer reviews it, considering not just does it work, but does it fit our standards, will it survive the next platform update. Only then is the proposal accepted.
What might have been an afternoon of trial-and-error becomes a few minutes of directed work.
4. BUILD AND DEPLOY
The developer asks the AI to confirm the deployment steps for the client’s hosting and CI pipeline, then executes them. The AI handles syntax and sequence. The human decides when to push and confirms the target environment. The irreversible decisions stay with the person.
5. QA
Finally, the developer has the AI draft a focused test plan: submit while logged out, submit on the protected page, confirm the entry lands and the notification fires. The human runs it and signs off.
THE DIVISION OF LABOR
Notice the pattern throughout. The AI provides speed, recall, and reach: instant research, request tracing, generated fixes, deployment steps, and test plans. The developer provides intention at every step: what to investigate, which answer is plausible, what constraints matter, when to deploy, and when the work is actually done.
The AI never decided anything that mattered. But it accelerated everything. Neither the human nor the machine would have reached this outcome as cleanly alone. And that’s the whole point.