Why fullstack architects matter more with AI

Nikos Katsikanis - March 19, 2026

Clients do not want to fund character by character software delivery

I still write code, but I think traditional hand coding is becoming a niche luxury for many client projects. Most clients do not want to pay several times more just so every line is written manually when AI can already produce a large share of the implementation faster.

That does not mean software has become cheap or trivial. It means the spending is shifting. Clients still need someone who can turn a fuzzy business need into a system that actually holds up in production.

I see my job as translation plus control

Most non-technical clients do not have the expertise to guide an AI toward a solid codebase. They know the business problem. I know how to translate that problem into instructions, boundaries, acceptance criteria, and architectural decisions.

That is why I increasingly see myself as the person who translates client requirements into prompts, issue breakdowns, and review rules for the AI. I also know how to reference the application code, data model, deployment pipeline, and infrastructure as code so the AI works with the grain of the system instead of against it.

AI is useful for project management as well as implementation

One of the most practical things I do is ask clients to put a new initiative into a large GitHub issue, Word document, or rough notes file. From there I can use AI to turn that source material into split GitHub issues, sensible labels, and an ordered task list much faster than I could do manually.

I touched on this more generally in Codex Tips, but I think the bigger point is managerial rather than purely technical. AI is not only speeding up code generation for me. It is also speeding up planning, decomposition, and backlog organisation.

I am still needed because AI can still drift badly

I do not see this as a future where the fullstack architect disappears. I see the opposite. I am still needed because it is still easy for AI to take a codebase in the wrong direction, especially when requirements are underspecified or the architecture already has sharp edges.

The analogy that makes the most sense to me is fly by wire. An unstable jet can perform extremely well, but only because the control system keeps correcting it in real time. AI assisted delivery feels similar. The raw capability is powerful, but it still needs constant expert correction to stay aligned with the intended trajectory.

The leverage is in architecture, review, and business alignment

When I work this way, my value is amplified. I am no longer weighted down by writing every feature character by character. I can spend more time on architecture, release safety, and making sure the software still reflects the client's actual business requirements.

This is also why I keep coming back to the practical AI workflows I described in Codex Tips and the commercial lessons in The hidden cost of short-term outsourcing. I use AI for acceleration, but I keep responsibility for technical direction, quality control, and commercial fit.

What I think clients are really buying now

I do not think clients are mainly buying typing speed anymore. I think they are buying judgment. They are buying someone who can understand the business, shape the backlog, keep the architecture coherent, and stop the AI from creating expensive chaos.

In that sense, I think skilled fullstack architects become more valuable, not less. AI increases my output, but the reason it works commercially is that I stay accountable for where the system is going.