A Lapsed Engineer's Return to Building

Slumber to shipping

The vibecoding renaissance is both a productivity story and a comeback story for people the discourse has largely ignored: the lapsed engineer.

These are professionals who trained in technical disciplines, grasped systems thinking and architecture, but at some point drifted away from hands-on building. Life happened, they took on leadership roles, had kids, and fell into commercial work that does not involve writing code. The muscles atrophied and every attempt to pick up Python again got cut short by some friction needing too much time and energy to navigate past.

For this cohort, agentic coding tools represent a genuine unlock, allowing them to reactivate dormant skills.

My personal experience is illustrative. An AI degree from the late 1990s, back in deep winter, followed by decades where the coding muscles went unused. Every sporadic attempt to bash out a bit of Python was a struggle. Then the vibecoding phenomenon arrived, and it felt purpose-made for someone carrying latent technical capability but lacking the time to stay at the leading edge.

The sunk cost trap

One of the harder realisations in any build is knowing when to start fresh. Lapsed engineers carry the baggage of half-finished projects, outdated codebases, and prototypes that got close but never shipped. A sunk cost fallacy keeps us tethered, while the thought of abandoning months of work feels wasteful.

The counterintuitive truth is that the tools moved faster than the projects ever did, and what took months to scaffold in 2025 now takes days in 2026. Starting fresh with modern tooling is often faster than dragging legacy work forward.

The Claude Code and Replit combination

The Claude Code and Replit duo exemplifies this. When I use them, version control is tight, the UX is consistent, deployment, which was once the obstacle that stalled many side projects, is no longer a hassle. When you reopen your computer, you do not spend fifteen minutes restarting the environment, it is all there.

This combination enables a different workflow, using Claude Code to interrogate the codebase, review architecture, and plan next steps. Replit then follows up with critique and further recommendations, before we eventually execute. What follows I've found in recent weeks to be sound, solid, reliable iteration that I can quickly and confidently push to live.

Why vibecoding suits experienced generalists

Vibecoding works best when you know what you want but lack the time to type it all out. That is the sweet spot for commercial leaders with engineering backgrounds. You understand architecture, trade-offs, and what good looks like. You can review and critique AI output. You just cannot spend every weekend grinding through boilerplate.

The caveats are real and vibecoding is not magic. I can believe the studies showing that while developers believe AI makes them faster, objective measurements sometimes show the opposite. There is a skills gap between prototyping and shipping production-grade software and experienced engineers will bring the technical expertise and critical eye that stops projects hitting issues of complexity, performance and security vulnerability.

But that is precisely the advantage lapsed engineers have. We know what questions to ask. We recognise when the AI is papering over cracks.

Building in public as the accelerant

Building in public is not new, but vibecoding turbocharges it. When you can ship meaningful progress in a weekend, you can narrate the journey in near real-time, bringing an audience along for the ride.

For lapsed engineers re-entering the building game, this matters. Building in public creates accountability and surfaces feedback early. It also builds a track record that might otherwise take years to establish.

Start iterating

If you trained as an engineer but stepped away, the tools have caught up to you. The skills you thought had atrophied are still there, they just needed a different interface. The vibecoding renaissance makes it possible for experienced generalists to build again.

Start fresh. Pick a modern stack. Ship something small. Iterate in public. The second act is available.