iiinova — where your digital self keeps evolving

Oh I don’t think it had access to the actual book content. But this one is so popular/buzzy right now I knew lots of people had been writing about it online, including spoilers. If I’d asked about some less popular, less reviewed book I likely wouldn’t have trusted the Claude report

040

Tangentially a very good use of agents! “Look up the twist ending. Don’t tell me what it is. But help me understand the nature of it: what level of gore? Any murder? Any chance I’ll have trouble sleeping after?” There was no way I was picking up a “thriller” without that filter.

280

…ruined the end for myself. But knowing the rough shape of how it ends makes me appreciate all the hints and subtle cues along the way of what’s to come. Doesn’t feel like I’ve ruined it at all. Makes me appreciate the story craft! The kind of book you could read twice and see new layers.

150

Maggie Appleton @maggieappleton.com · 01/05/2026

Obsessed with “Yesteryear”. So well written, funny, a necessary critique of the tradwives. I had to ask Claude for the gist of the twist at the end before reading, because it’s labelled a “thriller” and I am too soft to read anything truly horrifying. But it’s not that kind of thriller. Sort of…

The cover of the audiobook of Yesteryear on Spotify.

1130

Oh that’s good to hear the format works!

020

Maggie Appleton @maggieappleton.com · 27/04/2026

Forgive the ignorance, but what are managed agents?

120

Maggie Appleton @maggieappleton.com · 27/04/2026

How is everyone managing their agent SKILL.md files? Is it just chaos? Global skills, repo-specific skills, keeping them in sync between machines, figuring out which ones you have installed, authoring new ones. What are we doing?

30730

Maggie Appleton @maggieappleton.com · 23/04/2026

Not sure I understand - each session here is its own branch. And you can make forks off of each branch. Or roll back to a previous commit. It's all just git under the hood. Or did you mean something else?

Maggie Appleton @maggieappleton.com · 23/04/2026

Unfortunately agents all have *awful* design taste. Even when when supplied with detailed skill files on what I consider good design, there is still so much that can't be put into words: animations, interactions, taste, visual language. They faceplant in every category :(

230

Maggie Appleton @maggieappleton.com · 22/04/2026

Your new hobby seems to be being an extremely arrogant wanker to me and people in my replies. Will leave this long enough for you to see it and then ✨✨✨BLOCKED ✨ ✨✨

010

Maggie Appleton @maggieappleton.com · 22/04/2026

Extremely jelly of everyone who have a set of clear, measurable, computable metrics that define their tasks as “done” No 12 hour autonomous agent runs for designers who live in the tight feedback loop 😭

3570

Yeah this is a nice workaround. You can also put specs in issues and then point an agent at it to do a draft PR. But still hard to have real team discussions in the comments of a PR or issue. Too slow and async.

120

One Developer, Two Dozen Agents, Zero Alignment

You can read about why we built this and see lots more demos in the talk slides and transcript: maggieappleton.com/zero-alignment We're starting to roll this out to GitHub folks so we can squash bugs / make it less hacky. Then we'll do a public technical preview - there's a signup form in the link

maggieappleton.com

Why we need collaborative AI engineering

6332

Plans are also shared and editable as a team in real time. Because it is kind of insane to make elaborate plans with your coding agent alone in your terminal without getting anyone else's input. And then send it off to be implemented without checking if it's any good.

1120

It's multiplayer chats, plus agents, and everyone working in sandboxed vms so you're all sharing the same code and context. Prompt and code together, in real time. You're all connected to the same computer. You can all see the same outputs. No one is stashing work and switching git branches.

4201

...multiplayer humans all have to be in the same workspace. At @githubnext.com we've made a new research prototype called "Ace" in this direction It's like Slack, Github, Claude/Copilot, and a bunch of sandboxed cloud computers had a baby. Some demos:

2221

...communication that deeply integrate with coding agents Current versions of Github, Linear, Slack, etc. were built for a previous era of development. They're too isolated from agentic coding tools. Trying to duct tape coding agents onto them isn't working. Coding agents and realtime...

2131

Got to talk at @aidotengineer.bsky.social conf last week about the need for collaborative AI engineering. All our current coding agents are single player. We're trying to scale up individual productivity, but creating tons of alignment problems in the process. We have no good tools for...

Title slide saying "One Developer, Two Dozen Agents, Zero Alignment Why we Need Collaborative AI Engineering"

Slide saying "Alignment and communication is the new bottleneck"

Slide saying "Human context is not in the codebase"

Business context & financial resources, Political dynamics, Product vision, User research insights, and Organisational history are not in the codebase.

2213623

Lol okay yes Continuous AI is a different GitHub Next project. It’s really good! Customers love it. Sincerely. Genuinely *love* it. Has 80% WoW retention. Sounds like you don’t and that’s okay. Feel free to not use it Not what I’ll be talking about. But very proud of the rest of my team building it

010

Ah gotcha, looks like a separate research project happening at Microsoft / GitHub. It’s a big org. All kinds of research going on. Can’t speak to that one in particular. Not part of my talk. Showing a new prototype in this one focused on real-time human collaboration before agents touch anything.

Fudge bananas make mellow sounds with xylophones. Is a sentence that makes as much sense as your reply ☺️ In other words, what?

Yeah! Would love to come present and chat about it. Free whenever you want me.

020

Talks, workshops, events, and training for AI Engineers.

Yes! Will be on the AI engineer YouTube some weeks after: youtube.com/@aidotengine...

youtube.com

AI Engineer

020

Yeah current tooling is completely borked when you add agents to it. PRs showing all their cracks. Going to be a rough few years transitioning to the primitives that come after it.

010

Sincerely, what is “GenAI script”?

000

If you're at the conference, come say hello! I'm talking at 11am on Thursday, but totally free and much less stressed every hour following that. Will be riding the post-talk-stress-high and happy to hang out.

250

Maggie Appleton @maggieappleton.com · 26/03/2026

Putting together the final touches on slides for my talk at AI Engineer conf this Thursday @aidotengineer.bsky.social Talking about how we need realtime, multiplayer, collaborative agentic dev tools. Not isolated, single player terminal instances. And showing one we've built at @githubnext.com!

Title slide: One Developer, Two Dozen Agents, Zero Alignment: Why we Need Collaborative AI Engineering

The meme of two astronauts looking at earth. The one behind has a gun pointed at the other's head. First says "wait, alignment is the bottleneck?". Second says "always has been"

11485

After an unexpected flood of messages from my kid’s nursery about the “Easter bonnet competition” deadline, I have now learned this is a thing. Very camp. Very British. It’s giving Victorian children with more access to crafting supplies than expected.

A hat decorated to look like a yellow chick in grass

A set of colourful hats decorated with Easter eggs, flowers, grass, bunnies, and chicks

A hat with a woven bunny and eggs and carrots on it

1130

Maggie Appleton @maggieappleton.com · 25/03/2026

Oohh say more. I want this. How?

100

Maggie Appleton @maggieappleton.com · 22/03/2026

Whoaaa this is so much nicer than all the other floorplan software I had found! Everything else hadn't been updated since 2001 and was genuinely unusable. This is great, thanks

100

I’m also doing a floor plan in Figma separately, but couldn’t figure out how to integrate it. Mostly using that as reference and to plan furniture layouts. Figured it doesn’t matter too much as long as the key info is written somewhere in the obsidian vault

2110

Then told Opus 4.6 to design a system for me to manage within Obsidian. Said I wanted to manage: - budget and spending - essential house info - regular maintenance tasks - renovation projects - furniture research and purchases Told it to use Dataview and make dashboards for each main project

1181

I setup an obsidian vault and braindumped what I wanted into a plan.md file initially and the opened that in Claude Cowork. Messily listed out all the things I needed: projects to do, status of projects so far, contractors I’d already talked to, key house information, furniture to buy.

1140

Yeah Dataview makes all the good stuff possible! The best part is I didn’t even need to learn much about how it works or query lang. Just told Claude what I wanted and it made it happen.

160

Maggie Appleton @maggieappleton.com · 06/03/2026

Obsidian & Claude Cowork are doing all the house management and renovation tracking that I don't have brain space for right now. I ramble at Claude about things we need to fix in this very old and rundown Victorian we recent bought and it makes plans. Works great - both just operate on markdown.

A screenshot of Obsidian next to Claude Cowork. Obsidian has a bunch of files related to a home renovation. Claude is editing the files in Obsidian and adding tasks and plans.

71583

God grant me the wisdom to understand the internal logic of the credit system

Screenshot showing my credit score dropped this month because 1. I am not using my available credit, and 2. I’m using more of my available credit

6450

The looks exactly like a technical specification to me. Have you read the W3C specs? The source code is far more than 2000 lines long. So the spec is still an efficient, concise outline of the key architectural decisions you need to make source code.

IF you want to adapt software to fit your needs, run in your pref lang, change the architectural decisions, etc. rather than forking and editing source code, you can adjust at a higher level. Specs are a concise, directionally correct design ref with less complexity than the implementation code.

150

This is gathering chatter so I need to clarify: neither openAI nor I is suggesting everyone prompts their own version. Obvs wasteful. Theres a canonical implementation! The important line is “in your programming language of choice”. Where lang could be any number of design prefs and variables.

250

I should have put more emphasis on the “in your programming language of choice” part. Where lang can be swapped out for many design prefs The canonical code is one specific implementation. The spec-as-prompt means people have a higher level ref to make customisations to before they implement.

Agreed! Spec writing becomes the core process to build better tools for, skill-build around. Good product and design taste has always been the essence of great software, but now we get to put more energy and effort there.

040

The canonical implementation gets patched. And then they update the spec with the patch too. I should have emphasised more the “implement in your programming language of choice” part. Where “own lang” could be any variety of design preferences. So you only prompt bespoke versions when needed.

I would bet on this happening. Maybe models help develop programming languages that are much more efficient than human readable code. Or maybe we just use these existing langs? (Not my area of expertise)

000

Then I miscommunicated the general premise in my commentary. I very much meant to point to the highlighted part of the screenshot. The “implement in programming language of your choice” part. Alongside a canonical implementation. Where programming language could be any number of design variables

130

The spec is really good - detailed, comprehensive: github.com/openai/symph... Implementations are now easy. Writing excellent specifications becomes the core skill.

github.com

3391