You use AI as a chatbot. I use it as the ultimate learning engine. I break down hard technical concepts into simple mental models so you can build faster.
@jpwallhorn@cursor_ai Cursor’s training data set + RL pipeline gains are showing! Plus their partnership to unlock massive compute is going to help them push the limits.
Started using Composer 2.5, it’s quite good and the economics of it change the game.
Super excited to check out Composer 2.5 and what Cursor has cooked up. Was already loving Composer 2 and now getting near Opus 4.7 and GPT-5.5 level performance.
Introducing Composer 2.5, our most powerful model yet.
It's more intelligent, better at sustained work on long-running tasks, and more reliable at following complex instructions.
For the next week, we’re doubling the included usage of the model.
I can see this as the interface for work between AI agents and humans, but for transferring information between agents and sessions, I am not sure whether HTML is worth the extra weight. Markdown might still be better there in terms of token and context cost.
Using HTML to build the product spec contract and then converting it to Markdown is a pipeline I can see working as you start adding subagents.
From a business perspective, being able to self-optimize agentic loops. A "meta" cognition in the sense that it can suggest improvements to make to its own feedback loop.
For example, in code review, learn from a human who catches things that it did not, and figure out how to improve the agentic loop. What new data sources or actuators should be included to improve performance the next time around?
All that still feels fairly manual, and you have to really nudge in that direction.
"The spec is the product" -- Total agreement here.
Using the analogy of a coloring book, humans still define the lines (the spec) while AI can now color it in (proper shading and color theory), with a human verifying the final output.
Evals test the AI's ability to "color", and feedback loops help to teach it when things go wrong. Can remain agnostic on the tooling itself since they're improving so rapidly (Cursor, Claude Code, Codex, etc.) as long as the fundamental pieces are in place.
@ianlapham I'd usually be on the side of "why build it yourself when tools exist" but having a deep understanding of how your own system works unlocks something new and "meta".
Also, it seems you tend to stick to solutions you've made yourself rather than bouncing.
After 2 months of everyday use, I can say that setting up a personal research engine is one of the highest-ROI things you can do if you like to learn and stay on top of things at the edge
- Use a cloud-hosted agent, probably hermes or openclaw
- Learn about memory systems and encoding (cognee is very good at this)
- Build the right commands for parsing data and storing it (tag things properly, encode and save full text and key ideas)
- Build recurring jobs so the system grows itself (rss ingestion, auto twitter scroll, newsletter following)
- Build advanced skills that create connections between ideas, surface the most important info, and create digests for you automatically
- Build search retrieval skills that actually pull what you need and don't forget or miss things
Will change your life
@CWood_sdf Yeah, get the same vibes. Almost feels like we've reinvented coding from the ground up, just this time we do it in Markdown documents with English (which honestly might be the selling point?)
@nickco Clean up habits (e.g., alcohol, MJ, sleep) for cleaner, more stable energy. Unstable energy will lead to more issues downstream.
And then stack on top using AI to help separate signal from noise. Lot of noise so an AI is needed to help distill and filter the AI news haha
The question isn't 'will AI take my job?'
It's 'what is my actual irreplaceable value, and how do I 10x that with these tools?'
Answer that honestly and you'll never worry about AI again.
A 10x marketer with AI becomes a 100x marketer.
A commodity marketer with AI becomes...slightly faster at producing mediocre content at scale.
The gap doesn't close. It widens.
@tomfgoodwin Have centralized on Gemini. Too much negative press around ChatGPT and personally not trusting of Sam given the company’s history and ousting that happened.