In the past month I have shipped two social housing products to within touching distance of market, fixed a long-standing OneDrive backup bug in Swift, built a geographic system for identifying plots of land, written an agentic family-tree research application, brought delphi-dolphin.com back from a twenty-three year coma, and produced a narrated-walkthrough skill on a Thursday evening which now drives sales videos for any frontend I point it at. Seven artefacts, depending on how you count. One person. Evenings and weekends, mostly. The honest reaction is not pride at the volume. It is the quiet recognition I had forgotten what this felt like.
The spark, and what brought it back
I started my career writing software. Somewhere across two and a half decades of architecture reviews, governance frameworks, programme boards, and digital strategy decks, the writing-software part went quiet. Not absent, never absent, but parked. I missed it more than I admitted to myself. AI coding tools are the reason it came back. I am being deliberate about saying so directly, because a lot of writing about AI tooling treats the relationship as transactional. Mine is not. The tooling gave me my evenings back as a builder rather than a strategist, and I am grateful for it in a way no abstract benefits-realisation slide is going to capture.
The relevance to anyone reading this is not the gratitude itself. It is the fact the work happened at all. Seven artefacts in a month, none of them throwaway prototypes, all of them anchored in real problems. The question worth asking is what changed.
Cohesion is the design problem now
Building software used to be the hard part. Building it is now the easy part. The hard part is making the website, the writing, the video, and the social distribution feel like they came from the same person on the same Tuesday afternoon. Every layer of the stack is its own AI-flavoured tool with its own house style, and the path of least resistance is a portfolio drifting visibly between products because each product was generated by something different.
I treated cohesion as the spec, not the polish. One typeface system. One palette. One voice. A small set of named modes the work stays inside, no matter which tool produced the underlying asset. The frontend is the easy bit to get right because an AI coding agent follows a design brief faithfully when the brief is precise. The harder bit is dragging everything outside the frontend into the same gravitational field.
The pipeline matters more than the page
For the personal brand site, what sits behind the URL is a content pipeline doing the work I would otherwise have outsourced to four agencies in 2015. Posts originate as markdown drafts, get a structured editorial pass, render into the site, and feed a syndication step pushing carousels to LinkedIn through Publer.
None of these tools talk to each other natively. The integration is mine. The discipline is mine. What the AI tools accelerate is execution inside each step, not the design of the steps themselves. Anyone treating an AI tool as the pipeline is solving for the wrong layer.
The disposable-tool habit, and the moment it stops being disposable
The interesting move across all seven builds was not picking the components. It was building small, single-purpose scripts to glue them together and throwing them away once the job was done. A converter to push frontmatter from one format to another. A linter to catch the words I have banned myself from using. A git hook to stop me publishing anything still labelled as a draft. Each one took an afternoon. Each one would have been a week without an AI coding agent as a collaborator.
The exception is worth naming. On a Thursday evening I wrote a skill which takes a user persona and a job-to-be-done, reads the codebase to make sure it understands the current state of the frontend, drives Playwright to navigate the screens and complete the journey, calls OpenAI's text-to-speech models for the narration, and mixes the audio and video together with ffmpeg. The output is a sales-grade walkthrough video for whichever product I aim it at. With a little finessing, it works against any of my frontend systems. This is the inflection point where a disposable tool becomes capability. The skill is the artefact. The videos are exhaust.
This is the inversion surprising people who built websites in 1997. The cost of a piece of software has dropped to the point where the right answer to a small problem is to build a small tool, use it once, and bin it. The right answer to a recurring problem is to build a small tool, use it once, and keep it. The economic shape of the work is closer to writing memos than commissioning features.
What seven artefacts in a month proved
The volume is not the point. A solo operator, working in evenings and a couple of weekends, is now able to produce two near-market products, three working tools, a portable video-narration skill, and a resurrected personal site, with a tighter brand identity than most consultancy outputs I have seen this year. The artefacts render. The products work. The whole portfolio is held together by a handful of disposable scripts, one keeper, and a strict set of rules I wrote for myself.
The reason this matters beyond personal vanity is the same pattern applies to anything an organisation makes. Cohesion across products and channels is no longer a function of headcount or agency budget. It is a function of whether someone has thought about the rules holding the assets together and whether the disposable tools to enforce those rules exist. Most boards still believe they are buying capacity when they buy these capabilities. They are not. They are buying the rules and the discipline. Everything else is downstream.