Field Notes · Elle Anderson · andersonco.uk

Field note · AI native practice

How This Website Was Made

I designed and built this website myself. I started my career as a full stack developer, and I’ve stayed technical ever since, because I believe a product leader who understands the engineering adds the most value to a product organisation. I use AI the way a modern engineer does, as a tool that speeds up the work, while the thinking, the decisions and the standards stay mine. This page explains how it was made.

What I Set Out to Make

I wanted a site that would do several jobs at once: present a twenty-year product record to boards and executive search; explain a consulting practice to chief information and chief executive officers; give the product community free tools worth sharing; and demonstrate, rather than claim, that I build with AI as a matter of course. Those are different readers with different questions, so the first piece of work was not a page but a set of user journeys, one for each, and a decision about what each should find within two clicks.

How I Build

Every technical decision on this site is mine. I chose a static architecture served from a global content delivery network, so pages load quickly and there’s no server to secure. I designed the build scripts that generate the Console, its RSS feed and its search data from a single source file, and the serverless functions behind the live features, such as the weather ticker and the New Leaf coach. I also set the rules the automated checks enforce, from broken links to the house style.

I work in the code and the terminal every day. I write it, read it and review it, and I use an AI coding assistant for the repetitive work, as most engineering teams now do. Nothing goes live until I’ve read it, tested it in a real browser and understood exactly what it does.

Built With: HTML, CSS and JavaScript, written without a front-end framework. Python runs the build and quality scripts, and the content lives in JSON alongside schema.org structured data. Cloudflare Pages and Pages Functions host the site and its small API. A Cloudflare Worker runs the New Leaf coach, and Git and GitHub hold every change.

Tested With: Playwright, which drives a headless Chromium browser at desktop and phone widths and checks the printed PDFs. Python scripts check every link, the structured data and the house style, right down to Title Case and stranded prepositions. A weekly automated audit checks how search engines and AI assistants find the site.

Some of what that looked like on this site. I wrote the services and the track record in my own words and had them set, not rewritten. I rejected three versions of the homepage before the fourth. When a case study came back reading like a boast, I sent it back with the instruction that the piece was about children’s creativity and had to be written as such, and it was. When a field note used an analogy that was not mine, I replaced it with the Friday night in a pizza business that was. The voice guide in the brand file exists because I kept correcting the same faults until I had them written down.

What Was Built, and How

The site is fast, plain and built by hand, on one shared design system, so a change to any part is a change everywhere. Every printable, from the templates to the A3 chart of product lineage, comes from the same source as its web page, so print and screen never disagree.

New Leaf, where people create their squiggly career story supported by vibe coaching, is the most complete example. It assembles a professional story from a person’s own chapters, entirely in their browser, with an AI coach that follows tone and safety rules I wrote and keeps nothing. It went from an idea to a working first version in an evening, and that speed is the least interesting thing about it. What matters is that the product thinking came first: the opportunity, the users and their jobs, the strategy, the privacy and safeguarding position, and a log of every decision.

Alongside the site: a GitHub presence rebuilt from nothing, a pack of six product leadership skills that can be installed into any AI agent, and the Product Capability Compass, a private self-assessment for product people.

The Console is the newest addition. It holds a free guide called Computer Science For Product People, presented the way engineers see their own tools: a command line, a file tree, a scrolling ticker and function keys, with each visitor’s own date, time and local weather across the top. Eighteen modules explain the engine room in plain English, each with a diagram, a leadership view, something to try, further reading and questions to check what’s landed. Five Learning Paths bundle them for different journeys, and a dictionary answers any term a reader types. Alongside it sits LGTM, my weekly note on one technology story and what it means for product people, which readers can receive by email. The whole Console is built from a single source file, so a correction made in one place reaches the page, the feed and the search data together.

The Agents That Work for Me

I don’t only build with AI. I run a small team of AI agents, each with one job, clear limits and a human, me, accountable for what it does.

A findability agent runs every Monday morning. It checks how Google, ChatGPT, Claude and Copilot answer questions about me and my field, audits this site, fixes technical issues within strict rules and publishes the fixes. It also finds podcasts, publications and events worth my time, then logs everything and reports back. It never posts, contacts anyone, spends money or changes a setting. Those decisions stay mine.

A Console freshness agent runs every Wednesday. It checks four modules a week against the latest changes in technology, repairs any broken reading links and updates each module’s Last Updated date, so every module is reviewed roughly once a month.

A weekly LGTM agent writes each Monday’s note. It picks one real story from the week, checks every fact against at least two reputable sources and drafts the note in my voice. Before anything is published, a second, independent agent reads it as a future client or hiring manager would, checking that it is accurate, fair, kind and free of politics. Only a note that passes both checks goes live. Anything doubtful comes to me instead.

The Weekly Rhythm

• Monday, before the working day: the findability agent audits how search engines and AI assistants find the site, and fixes technical issues within strict rules.
• Monday morning: the LGTM agent researches, fact-checks and publishes the week’s note, after an independent reputation review, then leaves an email-ready copy in my drafts.
• Wednesday morning: the freshness agent reviews four Console modules against the latest changes in technology and updates their dates.
• Every publish: automated checks test the links, the structured data and the house style, and Git records every change so it can be reviewed or reversed.

Each agent has one job, written limits and a report it sends me after every run. None of them can spend money, contact anyone or change an account setting, and anything ambiguous comes to me to decide.

You’re part of the check too. Every Console module and LGTM note ends with a Report An Issue link. If you spot something wrong or out of date, tell me through the form. It comes straight to me, and corrections go into the next refresh.

Two AI coaches work with visitors directly. The product sense coach takes a real market or opportunity and asks one sharp question at a time. The New Leaf coach helps people with squiggly careers tell their story. Both follow tone and safety rules I wrote, and neither keeps what people type.

That’s the operating model I bring to clients: agents doing the repeatable work, people making the calls and the boundary between them written down.

The Standards I Set

Every page is checked in a real browser before it is published, at desktop and phone widths, and the printed output is checked as a PDF. Every link is verified after a change. Old addresses redirect rather than break. The palette, the type and the voice are written down and the scripts enforce them. Nothing goes live that I have not read.

What This Means for a Client

The method on this site is the method I bring to a product organisation adopting AI. The technology is quick and willing; the scarce things are judgement about what to build, standards for what good looks like and the discipline to keep the human responsible for the result. I have found that the leaders who get the most from AI are those who already know how to brief, to review and to say no with reasons, and that those skills can be taught. That is much of the purpose of my Private In-House Training for Product People and the operating model work.