It has been two years. The last post here was October 2024 and it ended with me promising photos. Well… clearly that didn’t happen.

Let me start where that left off, because I said in the second post that I don't like going back to edit, and I still don't. What follows is the catch-up, in four parts, one a week. This one is about where the two years went and what I decided at the end of them.

What happened to the Napoleonic campaign?

It did not start.

The 69th Foot were painted, the 69ième were painted, the Bourbonaise got their coats, and the first campaign, Sharp Practice on land and Black Seas at sea, never got as far as a first turn. Too many obstacles, as always.

Look at what I was promising myself in 2024. Bourbon French against Imperial French in 1807. The Ninth Legion sailing to Atalantes instead of vanishing in Caledonia. Russians against English in 1811 on plains where Europe's alliances don't reach. Every one of those needs the same things underneath it: a map with places on it, a history that says why anyone is there, a record of what has happened so far, and some way of working out what happens next when nobody is at the table. And before any of that, I had said I needed to start at 180 BC and work forward through AD 800, the 1630s and the 1750s to get to 1789 at all.

What I actually had was a borrowed map, a Copilot history I had called trash in print, and a notebook. The gap between the two was the obstacle, and it was not one more weekend's painting.

The figures did get painted though. That part of the hobby never stops. I will put a force in each of these posts, starting with the Tribal pre-Columbians below, because that is where the story has ended up.

The mounted warriors of the First Nations Atalanteans, with two totem poles.

The warband of Guards-the-bones on foot: Guards-the-bones in the tall feather headdress, the kneeling warriors, and the camp around its fire. Heroes behind, and his Warband behind them. Ready for the first battle of Alpha Campaign 1-1.

The gap

So what did I do with the two years? Mostly, I learned AI.

Properly, I mean, not the "ask Copilot for a blog post and laugh at the unit types" way of that last post. Although that post was not wrong about the method, and it is still the method: give it something, see what comes back, throw out the trash, keep the rest, change the prompt so it does not do that again.

In my day job I am basically an AI wrangler now, so while a steep learning curve, not as steep as hitting a cliff face. We all had GPT 3, 3.5 and 4 then Copilot, then for Atalantes I tried GitHub Copilot for actual building, not just prose, and it was nowhere near good enough for what I wanted. So I went to Claude, which is what I use at work, and moving to Claude Code was the best move ever. It is the most robust of the lot. That is the one I stuck with, and the one that has done the building since.

The thing I came to understand in those two years is that the obstacles that stopped the Napoleonic campaign were not painting obstacles or time obstacles. The obstacles were the kind of work that an assistant or a software dev team can do, if the assistant is good enough or you have the money for the software dev team and you are prepared to check everything they do. By the middle of 2026 I was ready to try again. My own Claude Subscription, my own plans, my own distilled desires.

This time the map is mine.

In September 2026 I sat down in Wonderdraft and drew a new continent.

Not the 2024 one. That was another designer's map, used with their blessing, and the only thing it shares with the current map is that both were made in Wonderdraft.

This one I drew myself, in two goes: the main island first, then three more large islands around it. I am still no cartographer. But I didn’t have to be. I drew the outline of the map landforms. Then started the geomorphology AI work. Step 1, place the island/s in the Atlantic. Step 2 AI.

What the AI work has done since is take my outline and enhance it, the coastline, the relief, the rivers, into something I would not be embarrassed to print at a decent size. Part 2 is about the map and the world, so I will stop here and just say that the shape is mine and the mountains are code.

The decision

AHere is the thing I decided, and it is the reason this blog is back.

I was never going to build the campaign setting I have wanted for thirty years by hand. Not the map, not two thousand years of history, not the tools that move armies about and work out who can see whom. I could pay someone to build it. I did the sums on that later, out of curiosity: for what now exists, a low-cost Australian development house would have wanted somewhere between $1.1 million and $1.9 million, or about $580,000 to $1 million from the cheapest kind of shop, the kind that hands the work offshore, with a team of seven taking ten to eighteen months. It is an estimate from the amount of work produced and the usual rules of thumb, not a quote, and the lower figure is the safer one. I will write that up properly in a later post, with how it was worked out, because the method matters more than the number.

What it actually cost was a monthly AI subscription, a few dollars of web hosting, and thirty-one days from the first commit to the end of the first development cycle.

So AI-built just became the best choice. Not the cheapest way to do it. The only way it was ever going to get done.

How it works, in a paragraph

I talk to one AI session, which I call the Conductor. It holds the whole programme, of work. When I say go, it writes a precise job for a separate worker session: what to build, which files it may touch, how it proves it worked. The worker does that one job, commits its work and reports back. The Conductor reads every change before anything is kept, and brings me the decisions. Most of my decisions are a choice between options it has set out, and quite often I take the one it recommends. Some I don't. I design, I decide, I check. I have defined the model to use and the effort the worker gives, and then Conductor starts it, lets it run, monitors what it outputs and reports back. Again a more in depth look in a later post.

The honest bit

It is not magic and it is not free of mistakes. On 4 October a one-character slip in the website's build file went live in one step and took the site down for about eighty minutes, with a damaged backup for company. Workers take shortcuts their briefs forbid, and say so in their reports, which is better than not saying so. Running several things at once ran the machine out of memory more than once. Each of those produced a rule: work lands on a development branch and the live site moves only on my word; every job carries an exact list of what it may do; one big test run at a time. The rules are the real product of the first month, as much as the code.

In that month, 1 October to 1 November, the count was 682 separate tasks handed to workers, 2,370 saved changes across six projects, and about 218,000 lines of software and tests. I have not read every one of those lines. I have read every report. And I was on holiday on a cruise ship in the Mediterranean from 5OCT to 24OCT and did all the interactions via my iPad.

What came out of September

The order of work was the one from the 2024 posts, just with help: first the map, then the structures that hold the setting, then the tools that do things with it. The map went into code on 21 September. The World Chronicle, the single record of everything that has happened in Atalantes, started on the 24th. The first rule set was chosen on the 26th. The Conductor took over on 1 October, and the first campaign was approved on the 4th. It was a very early release, it wasn’t horrible but it wasn’t something I’d do again.

The idea is the same as it was in the first post: a campaign world for historical war games, any period, where the battles have a reason and an outcome that matters to the next one. Same continent, same plan of starting at 180 BC and working forward. What changed is that the setting is now software, with a map, a chronicle, and a site where a campaign runs, and AI sessions do the drafting and the building to my direction.

Why not Napoleonic

One more thing before I stop, because it is the question anyone who read the 2024 posts will ask. The first campaign is not Napoleonic. It is tribal, set in the summer of 131 BC, fought with Tribal, the skirmish rules from Mana Press.

Partly that is the history: it starts at 180 BC, so the first campaign starts where the history starts. Mostly it is simpler than that. I had played a couple of games of Tribal, a friend was interested, and Napoleonic games are hard to find locally. I built the New Recruit force files for Tribal, which got me talking to Ara Harwood, the designer from Mana Press. So it just worked. There may be a post on the Tribal New Recruit files at some point.