Richard MacManus

After my trip to the @aiDotEngineer World's Fair in San Francisco earlier this month, I wrote up the 5 key learnings I took from it. Here's a short thread of each of the trends dominating AI engineering now. Check out the full post on @latentspacepod. latent.space/p/aiewf26trends

Loops, harnesses, factories and why engineers need to own the outer loop.

Turns out that the liberal arts education was useful after all. When anyone can author a dashboard or new web application with a quick prompt, a broad exposure to different schools of thought helps you know what's worth building.

In Defense of Writing. A disturbing trend I see emerging is people lazily generating longer form content from bullet points or dictation they give to an LLM, and then passing it off as their own. Beyond the obvious breach of trust and social expectation when doing this, there are several more nuanced but equally insidious problems with doing this. First, writing is thinking. If you aren't taking the time to write, edit, rework, reshape, and play with your words, you're shortcutting a key aspect of refining an thought. You rob yourself of this opportunity if you just skip to the end. I have yet to read anything generated that even comes close to the clarity and impact of a well-written piece. Second, respect. Very few people will tell you this to your face, but as soon as they read something 'from you' that has all the tell-tale signs of AI slop, their respect (and trust) for you drops. What you are communicating is that you don't take the time and effort to do things properly, or at least entirely honestly. This is damaging in way many don't realize. I instantly discount anything I read as soon as I recognize it as AI slop. Third, and I think most important, you are training yourself to take the easy way out. I think this is different than 'working smart'. It's intellectual laziness, effectively choosing cheap calories rather than a nourishing meal, even if it takes longer to prepare. That effort is where good things, as well as your own growth, come from. Many days on here I feel like the last man alive who wants to actually read what you wrote. It's entirely possible that this is a losing battle and the nature of media and medium is trending in an irreversible way. But I really hope it's not. I'll leave you with a philosophical question: every social media platform is going to great lengths to label AI-generated images and video. Why not writing, as well?

.@geoffreylitt believes understanding the work that AI creates helps humans participate and compound in their efforts. He spoke about this at @aiDotEngineer recently and shares how his inspiration of education led him to build explainers that deepen his understanding while still pushing the frontier.

Since you can see me now, here's a site I made to demo/explain WebMCP: Maybe you've heard about WebMCP, but if you'd like to see it in action and learn how to work with it, how agents use it, what the spec looks like, and some important links? It's interactive! Use the extension to actuate the page:

This week's Agentic Web News covers the rise of the agentic web as a concept, browsers in agents, real momentum for the WebMCP protocol, & more! agenticweb.news/browsers-in-ag… Mentioned: @liadyosef @kieranklaassen @i_am_kaspa @jason_mayes @andreban @andyfang @Runtypical @nekuda_ai

Thrilled to see my first @aiDotEngineer talk featured in @latentspacepod’s five trends defining AI engineering—alongside @romainhuet, @trq21, Pauline Brunet, @_philschmid, & @garrytan. @aidotengineer has long been my favorite source for substantive AI thinking—so this is very cool and borderline surreal! In “WTF Is the Context Layer?” I explain what a context layer is, why it matters, and which approaches I believe will win.

This afternoon I spent time with the new Codex/ChatGPT, as I will use it a lot more going forward. I started with a Mastodon bot I run, Classic Web, which posts screenshots from old websites. I refactored it using Codex, removing some of the manual parts. Still testing, but wow.

Vint Cerf now advising Innovation Labs, "an organization trying to create the open architecture for AI agents to identify themselves." (1) I interviewed this co recently (2); identity a key piece for agentic web. 1. techcrunch.com/2026/07/15/vin… 2. agenticweb.news/who-owns-ai-ag…

After my trip to the @aiDotEngineer World's Fair in San Francisco earlier this month, I wrote up the 5 key learnings I took from it. Here's a short thread of each of the trends dominating AI engineering now. Check out the full post on @latentspacepod. latent.space/p/aiewf26trends

The term “AI engineer” was introduced by @swyx 3 years ago. It started life as a new job title, but at AIEWF 2026 it felt more like a description of where software engineering itself is heading. Lots of learnings ahead, what an exciting time to be a dev! latent.space/p/aiewf26trends

5: Every agent platform is building around skills. If "loops" was the main buzzword at AIEWF, "skills" was a close second. There was excellent practical advice from @_philschmid, @pbakaus, @mattpocockuk; and some inspiration from @garrytan.

🆕 5 Trends That Defined AI Engineering at World’s Fair 2026 latent.space/p/aiewf26trends @ricmac's big recap of @aidotengineer: 1. The focus shifts from agents to systems 2. Loop engineering is the new control layer 3. Al engineering enters the enterprise 4. Coding agents replace IDEs 5. Every agent platform is building around skills

MCP is quickly turning into the HTTP of agents. If I were building a site tomorrow I would build it MCP-first, because that is how agents increasingly want to consume software. Exposing those capabilities without governance is a liability though. Got into all of this with @ricmac: agenticweb.news/arcade-actions…

My personal reflections on a hectic and inspiring week in San Francisco, covering the AI Engineer World's Fair. agenticweb.news/ai-engineer-wo…