Own the Why, Rent the How: How I Actually Use AI to Get Work Done

I have more ideas than I will ever build. For years that felt like a personal failing. Now it just feels like a supply problem, and I have finally found the machinery to deal with it.

Here is the honest version. I know basic HTML, CSS and JavaScript, and lately a bit of Flutter and Dart. That is it. Not long ago that would have been the ceiling on everything I could make. It isn't any more, and the reason isn't that I got better at code. I got better at delegating it.

The Firehose

My problem was never a shortage of ideas. It's the opposite. They arrive faster than any human could act on them, most of them rubbish, a few of them gold, and no obvious way to tell which is which at 11pm on a Tuesday.

So I stuck an LLM in front of the firehose and told it to be a bouncer. Every scrappy idea goes in. It pressure-tests them, groups them, ranks them against what I'm actually trying to do. Chaos in, structure out. Then I run the same filter on the way back, because a model will happily hand you ten mediocre options dressed up as ten brilliant ones, and my job is to bin nine of them.

The real win isn't speed. It's that I have an ADHD brain, and focus is the scarce resource, not time. Every boring mechanical step I hand off is willpower I get to keep for the part that actually needs me.

I Own the Why, I Rent the How

Think of it like running a small film set. I'm not the one operating the camera or hammering the set together. I'm the one who decides what the film is about and why anyone should care. The crew handles the how.

That's the whole trick. I decide what should exist and why it matters. The coding agent handles the syntax, the scaffolding, the fiddly bits I'd otherwise be googling for an hour. I haven't stopped learning - I've just stopped letting my typing speed set the ceiling on what I can ship.

The bit that makes it repeatable is writing things down first. Before I build anything I write a prd.md (what and why) and a design.md (how it should look and behave). The agent builds against the spec instead of my vibes, which means fewer wrong turns and a document I can hand to the next agent - or the next me - without re-explaining the whole thing from scratch.

My Crew of Little Robots

At work this stops being a hobby and becomes how I operate. Rovo, Atlassian's agent layer, is the one that changed things, because it isn't a generic chatbot bolted on the side. It knows my tickets, my docs, my project. Context is the whole difference between a party trick and a tool you actually reach for.

So I built a crew. Each one does a single boring job, properly. And because every one of them is just a skill file on disk, here is the actual brief each robot runs on - straight from its SKILL.md:

Sharkitect - takes a rough idea and shapes it into a proper Jira Epic, Story or Spike against a clean template, walking me through the sections I'd otherwise skip.

Build and refine high-quality Jira tickets — Epics, Stories and Spikes — by applying the correct team template, guiding the user through every required section, and pulling supporting context from related Jira tickets, Confluence pages, Miro boards and GitHub code.

Beacon - pulls a JPD Idea or Epic and drafts a test plan straight from it.

Meeting Notes Formatter - takes my scrappy live scribbles and reorganises them under a fixed set of headings, without inventing anything I didn't say.

Turn unstructured or scrappy meeting notes into a clean, well-formatted document under a fixed set of headings — context, decisions, scope changes, risks, next steps and Q&A.

UNO Reverse - reads the internal compliance checker's own guidance, then redrafts my ticket to score highly against it. Fighting the graders with their own rubric.

Raise a Jira ticket's compliance score by reviewing the CheckMate / Risk Review documentation and the ticket's comments — especially automated Checkmate comments flagging gaps — and producing a revised draft of the ticket.

Product to Design Brief - turns a finished product doc into a structured UX/UI brief, either for a human designer or ready to paste straight into an AI design tool.

Transform a PDD, PRD or product idea into a structured UX/UI design brief following the NewDay Cards template — either as a copy-paste-ready prompt for an AI design tool, or as a detailed brief for a human designer.

And a second bench for the truly thankless jobs:

A PDD Generator that turns scrappy notes into a proper product doc:

Turn scrappy notes, chat messages, meeting notes or Jira issues into a standardised Product Definition Document (PDD), or into the initial Idea template for Deko Technical Roadmap items.

A RAG Syncer that keeps my Ideas and Epics telling the same story in the right tone for each audience:

Synchronise RAG (Red/Amber/Green) status and commentary between linked Jira Product Discovery Ideas and Jira Software Epics, re-toning the commentary for each audience — delivery detail going up to Epics, executive summary going out to Ideas.

An Acceptance Criteria tidier:

Find acceptance criteria buried in a Jira ticket's description and move them into the ticket's dedicated Acceptance Criteria field, tidying up the leftover heading and duplicated text.

A Technical Ticket Translator that turns engineer-speak into something a stakeholder can read:

Translate technical Jira tickets into plain-English summaries for Product Owners and non-technical stakeholders, covering the problem, the impact and what success looks like.

And a Release Note Writer for release day:

Generate concise, user-focused release notes on release day for Jira tickets labelled "rnw" (release note worthy), categorised by journey, written back to the ticket's Release Notes field with alternative drafts offered in a comment.

GitHub Copilot handles the smaller ritual - spinning up my daily note, applying the template, tidying the mess.

None of these are clever. That's the point. They're the ten-minute jobs I used to put off until they became forty-minute jobs.

All of these skills can be grabbed from here: https://github.com/nbinding/nb-skills

The Bit I Didn't Write

Here's the part that tells you where this is heading. I didn't write this post from a blank page either. There's a skill sitting on my machine that knows my voice - the spaced hyphens, the British spelling, the habit of ending on a question - and it drafted this while I supplied the opinion and the edits.

Which is the same deal as everything else here. I brought the why. Something else filled in the how.

So I'll leave it with you: if the machine can draft the ticket, write the spec, tidy the notes and even mimic the way I write, what's actually left that only you can do - and are you spending your best hours on it, or on the busywork a robot would happily take off your hands?