Tim

Finally finished a little side project I've been working on! An Intel 8080 Emulator written in Swift that emulates the original arcade version of Space Invaders! You can find the emulator here: github.com/timsearle/emul… And the game here: github.com/timsearle/spac…

I give it 12 months for folks to realize that they e simply forgotten about what makes a product good. When you move so fast that everything is a blur, you don’t realize that you’re driving in the wrong direction.

@rv_RAJvishnu @GergelyOrosz Copilot CLI does all of this (it’s a fork of Codex)

the jump from autocomplete to full agentic iteration is happening way faster than most realize... copilot is great for syntax but it doesn't grasp the architectural context that claude code or codex can handle. once you get a taste of an agent that actually runs tests and fixes them in real time, going back to simple suggestions feels like a massive downgrade. the licenses are being canceled because the workflow changed, not because the tool failed

Actually, I do hear more startups “taking away” GitHub Copilot from devs - and no one is complaining at those places. Because those devs don’t use Copilot, and are on tools like Claude Code, Codex, Cursor agents etc. So companies just cancel the unused Copilot licenses.

@GergelyOrosz Copilot CLI seems to be incredibly under utilised/marketed. Their USP allows you to use many different frontier models under the same license. The harness is definitely not as good as Claude Code et al, but being able to use all the frontier models via CLI was a great feature.

@andrarchy It was more an experiment/POC utilising existing technologies rather than everything needing to be something new. Need to figure out the balance around approvals vs audit vs TTL. But yeah - if we want to allow an assistant full access, there’s risks to consider. All a trade off

@timsearle_ Tbh cure seems worse than the disease

This simple comment can save you weeks of work trying to get OpenClaw to follow your rules. I know because I spent those weeks writing validation scripts, building approval flows from scratch, and adding rules to my agent's docs hoping it would follow them. My agent was supposed to get approval before sending messages. It didn't. It was supposed to validate schedules before creating cron jobs. It skipped that too. So I started building what I called "enforcement wrappers" — custom scripts that check the rules before letting the agent act. It worked, but it was a lot of effort and now I had a growing pile of one-off scripts, each solving one problem. Then I found out someone had already solved this. Not only that, IT WAS ALREADY IN OPENCLAW. It's called Lobster. So this is a great example of two things. 1. If you really need something in OpenClaw, there’s a good chance @steipete and the other contributors have thought of and put it in there. 2. Just because OpenClaw has a feature, that doesn’t mean your agents will automatically use them. So what is Lobster? Lobster lets you define workflows where every step has to complete before the next one runs. You can require approval at any point. If approval doesn't come, the rest of the workflow doesn't happen. Simple as that. You enable it with one line in your config and start writing workflow files. Here are some real-world examples of how we're using it: Sending messages to real people. My agent drafts messages to my wife, clients, collaborators. With Lobster, the workflow has a built-in approval gate. The message literally cannot send until I approve it. Creating project plans. By default I want every plan to have at least three milestones and one goal. The workflow validates before it writes anything. No milestones, the workflow prompts my agent to create them. All plans always get milestones and goals (technically OKRs). Scheduling automation. This is a big one. My agent creates cron jobs for things like email checks, PR monitoring, security scans. It used to pile them on top of each other at round minutes, causing jobs to collide and fail silently. The Lobster workflow validates the schedule against existing jobs before creating anything. If you're running OpenClaw and you've been writing custom scripts to keep your agent in line, check out Lobster before you write another one. Docs: docs.openclaw.ai/tools/lobster

@GergelyOrosz @frantzfries The market will select for it. If it’s a problem, those people will get the less favourable jobs. If it’s a non-issue, things will continue as they are - bugs happen, they get fixed.

@frantzfries gotcha. yeah I hear you I wonder how this will be different... will it be sustainable? I DO remember when I started my first job seniors thought those using autocomplete would honest to god be terrible devs BUT AI allows you to turn your brain off, true (if it's how you do it)

Same version of this from ~20 years back: "I swear to god the average new grad is Google-pilled to the point of being unable to complete basic tasks on their own, without searching online" (prevailing wisdom back then was that real devs know stuff by heart or use books)

I track my workouts in text form in Obsidian. So I have total flexibility and speed, then I paste into the claw at the end which normalises it and stores it against a schema, which is then used to power a local web app that visualises all my exercises, progress and goals. For me this is the perfect combination of optimised input vs structured analysis. Taking photos of your weights each set is hilarious 😂

this is so much more work than like 2 taps to log in literally any workout app ever made. also doesn't show rest time. have to mentally keep track of sets or ask it or scroll up. have to ask it what you did last time instead of just a 3 second glance. what are we doing

@summeryue0 x.com/timsearle_/sta…

Nothing humbles you like telling your OpenClaw “confirm before acting” and watching it speedrun deleting your inbox. I couldn’t stop it from my phone. I had to RUN to my Mac mini like I was defusing a bomb.

Written more about this authentication and authorization flow here: searle.dev/2026/02/15/age…

For my @openclaw setup, I’ve introduced what I call the “Clawditor”. A separate channel that notifies me with an audit event when my bot performs sensitive tasks, and requests approval through back channel authentication for even more sensitive tasks. This allows me to separate read vs write access tasks on my email, no need to trust the AI here. You do lose some flexibility and have a risk of approval fatigue, though. But it’s more transparent.

@GergelyOrosz Obviously the people generating this stuff have other motives, but I’m trying to use AI an absolute minimum when I write my blogs. I use them to think. This article really summarised why for me cassidoo.co/post/good-brai…

I am getting fatigue from reading more and more AI-written articles on this platform I get a sense that it’s AI-written a paragraph or two in, and full certainty by paragraphs 3 or 4. Something about the cliche phrases AI loves to write, the neat and utterly predictable structures, the overused terms…

God, I detest this dumb argument every time it rears its ugly head. "Your brain is doing the same thing as an LLM!" Um ... no? When I write a poem about how a strawberry tastes, I've actually tasted a strawberry. The LLM hasn't, and in fact can't conceptualise taste at all.

📣

@DonnyWals Would love to try it and give some feedback 👌 I’ve not tried much with Apple’s on device models myself yet, so I wasn’t even if what I’m describing is possible. It just took me 10 years to realise I didn’t need to re-invent the notebook, but I did want data visualisations 😂

The “main” flow I use is to track without a plan and that gets relatively close to Freeform tracking where you can just add the exercise and either log three sets at ones or go one by one for example. I’ve though about a notes -> structured feature but I’m not sure I could make it fit. If you want, I can get you on the TestFlight so you can give it a go for a few workouts, see if you like it

One of the funnest things about building Maxine is that I'm just learning so much new stuff about strength training and how people like to use apps while they do it. My small TestFlight group is giving loads of feedback and it's all so incredibly useful

@DonnyWals I basically want the best of both worlds - text input, but all the benefits of it being stored in a structured format for the data views. I just hate UI for the input.

@DonnyWals I’ve always struggled with fitness apps and their UX. Recently my workflow is OpenClaw to convert my rough gym notes in telegram into structured data - so I’m not encumbered by UI. Maxine looks great - I’d be keen on an Apple Intelligence feature in it that achieved the same? 👀

@lexfridman Spot on, your intuition is right. But I also don’t think this is all on the builders of the assistants. There’s a huge responsibility on the services they interact with to provide better authorization models. I wrote up my idea here: searle.dev/2026/02/15/age… POC soon

The power of AI agents comes from: 1. intelligence of the underlying model 2. how much access you give it to all your data 3. how much freedom & power you give it to act on your behalf I think for 2 & 3, security is the biggest problem. And very soon, if not already, security will become THE bottleneck for effectiveness and usefulness of AI agents as a whole (1-3), since intelligence is still rapidly scaling and is no-longer an obvious bottleneck for many use-cases. The more data & control you give to the AI agent: (A) the more it can help you AND (B) the more it can hurt you. A lot of tech-savvy folks are in yolo mode right now and optimizing for the former (A - usefulness) over the the latter (B - pain of cyber attacks, leaked data, etc). I think solving the AI agent security problem is the big blocker for broad adoption. And of course, this is a specific near-term instance of the broader AI safety problem. All that said, this is a super exciting time to be alive for developers. I constantly have agent loops running on programming & non-programming tasks. I'm actively using Claude Code, Codex, Cursor, and very carefully experimenting with OpenClaw. The only down-side is lack of sleep, and an anxious feeling that everyone feels of always being behind of latest state-of-the-art. But other than that, I'm walking around with a big smile on my face, loving life 🔥❤️ PS: By the way, if your intuition about any of the above is different, please lay out your thoughts on it. And if there are cool projects/approaches I should check out, let me know. I'm in full explore/experiment mode.

Can’t believe how much has changed since the first London Claude Code Anonymous meet-up back in August…! Wild times. Congratulations Peter.

Peter Steinberger is joining OpenAI to drive the next generation of personal agents. He is a genius with a lot of amazing ideas about the future of very smart agents interacting with each other to do very useful things for people. We expect this will quickly become core to our product offerings. OpenClaw will live in a foundation as an open source project that OpenAI will continue to support. The future is going to be extremely multi-agent and it's important to us to support open source as part of that.

@DonnyWals Yeah - the personal assistant/chatbot is the ultimate interface. Siri still required people to understand the shortcuts app, automations, and advanced features of iOS, and on top of that, developers had to consistently support it.

@timsearle_ Yepppp same. I kind of gave up on this dream but I think with ai and agents becoming more important, we might be on track again?

"Apps are dead" I've been thinking about this to some extent since roughly iOS 10 which is when Apple allowed apps to take up space outside of users launching apps. (notification extensions, spotlight indexing etc) At this point, users can pretty much interact with everything in your app without ever launching it through App Intents. It's really not that crazy to think that, if Siri gets better over time, natural language will translate into App Intents, and your app will receive requests without the user actively having to name or launch your app. I doubt apps will die anytime soon. But I do think that we're entering an era where integration with the OS will make all the difference between an old fashioned "must open to use" and a modern "just works like magic" app. Design for this and you'll have an edge over competitors by the end of the year