Context Inertia is when an agent keeps chasing a goal after the evidence says it should stop. Persistent, long-running models make it worse. Compaction won't fix it.

I think there are two ways in which people have been designing systems. The first approach is that you understand what the requirement is and then you try to build individual components to meet the…

ai kills the idea maze and the stress tests that happen in the 0 to 1 process the idea maze gets built and refined subconsciously when you write the prd, the deck, the 1 page north star doc, the…

A discussion of the OpenAI release of 372 solutions to math problem and its implications, through an analogy with the abelian sandpile model

This article is available as a video on the Modern Software Engineering channel. As a technical coach, I find Mob Programming is really effective. Recently of course, we started using agentic AI, and…

For simplicity's sake, I use the term "AI" to refer to LLMs and other related technologies throughout this post. None of these things are actually AI. Also, I ascribe human-like characteristics to AI…

Five checks for shipping agent-written code, with a 5-minute self-test. Read it here or download the free PDF. The post How do you trust AI code without reading it? appeared first on Renato Budinich.

So, I think "intense ambivalence" - in the sense of being pulled in two directions - is the most reasonable place to be about this whole AI thing. That's kind of a disclaimer - I don't think AI is…

There are some simple tests to determine whether Amodei (and Altman, and Musk, and Zuckerberg, and Nadella, and Pichai) are serious about any commitment to safety, alignment, “pacing the…

Will Our Writing Survive AI? Surviving AI Contamination Substack post from @islandsofthefuture, the Substack I started for my novel in progress. I feel I am being contaminated by AI. Two years ago, I…

Credit Where Credit is Due October 14, 2026 This summer, Santiago Schnell, the provost of Dartmouth college, wrote an article for The Washington Post titled Universities are fighting AI cheating. But…

I tried agents with skills. When they failed, I found swamp.

Ask a coding agent for a cache in a particular language and watch what it does. It may go looking for a library. But I often see mine simply build one for the task at hand. No comparison of five…

A few years before the advent of gpt-3, a colleague introduced me to an entrepreneur who had a problem he wanted to solve. He ran a “human supermarket shelf supervisors” business where…

As Swamp Club launches and I blow out candles, we can start to look towards a new future of Agentic Ops.

The ultimate product forms of artificial intelligence can be summarized as the following: Virtual employees: take on clearly defined roles, execute tasks continuously, and report to supervisors.…

AI is not only for the tech folks. That was the whole idea behind my session with the members of Rajkot Management Association Business owners and individual experts, mostly from non-technical…

So we’ve entered the era of personal AI agents as assistants. The most famous one being, of course, Meta’s new assistant Muse. What’s the obsession with personal assistants?…

AI can now generate clean, well-tested code in minutes, but who ensures we're building the right thing? This article explores why engineering judgment, challenging assumptions, and understanding…

It's a sensitive subject The post Asking ChatGPT to be my life coach appeared first on Aman Y. Agarwal.

If you’re active in AI or software development circles, you’ve probably heard about Jev in the last few weeks. It seems like everyone, myself included, is excited by the possibilities of Jev and the…

System One decision models fit on-device behavior detection, and Chrome is building one into the browser. For real-time protection, small models trained for the task still win on latency and memory.

This blog post summarizes our paper which will appear at the Transitioning from Pre-Training to Post-Training workshop at NeurIPS 2026. This is joint work with Ashutosh Sharma, Avinash Balakrishnan,…

Is AI helping leaders think better, or helping them avoid thinking? In this episode of Creating Value Wins, Johan Roos […] The post 260430 Is AI Making Leaders Think Less with Johan Roos |…

A look into how agents in a simulated environment coordinate and adapt in order to complete a shared long-horizon task.

Claude Code can already consult a bigger model, but only Claude's, when Claude decides, reading my whole conversation, and with no record of whether it paid. So I built a panel that asks two vendors…

AI decision models answer typed questions in milliseconds for a fraction of LLM cost. Where Clef-Flash shines, where it fails, plus a live demo.

Coding agents resend full context every turn and providers cache by exact prefix match. See how each harness shapes that prefix and why the economics differ.

We all can learn something from the latest episode of the OpenAI drama, this time involving corporate espionage, a foreigner, and an absolute slop attempt to steal insider information.

My favorite thing about Swamp is not another agent. It is having reusable pieces that let me solve the next problem without asking an agent to rebuild the last solution.

Filming a trailer for TrustMe, almost fully run by Astra xhigh - with special help from Opus.

People keep asking me how I do agentic data science with marimo notebooks. Recently I stood at a whiteboard and drew the whole system out: the setup, the tools, the loop you work in with the agent.…

1:00 pm. "ok, awesome, clear the notebook of any code + commentary cells, we want to return it to the state where there are just hints." 1:02 pm. "clear out the slide deck too." 1:03 pm. "clear out…

The Emperor Pelican Has No ClothesOctober 14, 2026 Why do all LLM pelicans look like this?When given the prompt Generate an SVG of a pelican riding a bicycle.Original SVGs from Simon Willison’s runs…

Reading Stuart Jeffries on the Frankfurt School, the 1920s–30s diagnoses of mass culture and alienated labor sound uncomfortably like descriptions of AI now.

Moll and Imas argue that AI is unlikely to cause rapid growth in the next ten to fifteen years. Here is why I wasn't persuaded.

My experience with multi-agent workflows in daily work. The agents keep each other's contexts clean and on track, and my job has become what I would call context juggling.

The anime-only episode where Chii buys panties is a rare improvement over the manga As I wrote about in my previous blog post, I read the Chobits manga for the first time after rooting an Android…

Me, calmly assessing the implications of LLMs for programming language design(ers). As someone who has made a career of programming language design, I definitely have moments of existential dread…

A single live debugging run through a multi-agent incident response system: how a shared incident board, per-agent tool scoping, evidence minted by tools, and a deterministic gate catch a silent…

Code here In my last post, I trained a world model that attempts to play Pokémon Red. The model learned to do this through its latent space, and used the learned predictions to plan actions in the…

Reflections on using Grok Bot for a week, exploring its potential and challenges as a solo entrepreneur.

An agent adds an address to a production allowlist at 2:13 a.m. The dashboard says approved and succeeded. Which system saw the change, and who decides whether that record is enough?

What's the "100ms threshold" for AI UX?

How to (not) vibe-code at work My last AI journal entry was about vibe-coding at home for fun. I still do that. A lot. I will write more about that later. At work, things have changed as well. Some…

I had agents write up sixteen of my builds and gave other agents the sources and a veto. All sixteen came back corrected, and the errors leaned toward what a reader would hope was true.

I've previously shared my vision that the industry is moving towards software factories for 99% of code changes and to make that effective we'll need to move to a paradigm of Durable Context and…