Our workflows are relatively simple, so usually the issue here is large files associated with the workflows, which are we’ve mostly solved by using subagents for file processing.
Beyond that our compaction strategy is literally “use a leaked version of Claude code’s compaction prompt” after 80%
Wrote up a series of nine short posts on interesting problems we encountered while building an internal agent framework (virtual files, compaction, LLM vs code-driven workflows, etc)
lethain.com/agents-series/
Great site! It’s pretty cool how approachable projects like this are getting
Thank you! 🙏
There's also a sense, for me anyway, that the frontier has moved somewhere else -- I'm not quite sure where -- but a lot of social media today feels like a regurgitation of things that already happened rather than something new/interesting/evolutionary.
Yeah, absolutely an era I miss fondly. There was a post from @copyconstruct.bsky.social a week or two ago that captured the reality for me -- most of the folks I want to talk with are gone, and I've personally *mostly* left social media for all significant purposes.
Digital edition of my fourth book, Crafting Engineering Strategy, is up for purchase! (Print copy should start shipping in next 2-3 weeks.) Tried to apply "Staff Engineer" formula to eng strategy topic, with a lot more editing 😂
www.amazon.com/Crafting-Eng...
*Good 😅
Could point! This probably does exist somehow already
Yeah, so you can imagine another stage in this process of summarizing all rfcs to extract problems, but I’m more specifically looking for a pattern we could scale across many functions, eg compliance feedback, security feedback, etc
Any good patterns for loading a Notion page as modifiable prompt to respond to other Notion pages and Slack messages? Eg as a simple example you could imagine I want to write architecture guidelines and then use that to comment on RFCs.
Is this just Zapier?
I love your post!
New post up, with a bunch of thoughts on what drives me personally, and my own 40-year career (h/t @lethain.com).
I have a pretty unusual life story. But one of the reasons I love the tech industry so much is how common unusual stories like mine have historically been.
charity.wtf/2025/07/09/t...
Did my first podcast in some time as I start ramping up a bit around getting started as new CTO at Imprint and also getting close to releasing next book!
Podcast up at www.executiveruntime.show/episodes/04-...
I think programming languages are already pretty exhaustively represented in the training sets, although it certainly varies language by language, and providing more detailed in-context learning is usually preferable in my experience.
Have six (draft, unedited) chapters of my upcoming book exploring this idea in a bit more detail!
craftingengstrategy.com/aic/generate...
Full piece at lethain.com/competitive-... going into the concept a bit more deeply.
Over the last year, as I've spent more time working with LLMs to power product features, I've also been thinking about how LLMs impact me as a long-form author. Of the ideas I've had, I'm particularly interested in the idea of building, and selling, books as "datapacks" for LLMs.
Part of this project that I'm excited about is also releasing an LLM-optimized version of the book designed to be throw into an Claude/ChatGPT project as in-context learning, which I'm working with O'Reilly to distribute. Not available yet, but you ca see use examples on website!
You can already read the full, unedited version of the book online at craftingengstrategy.com (most of those have also come out on my blog over the past 18 months as I wrote)
Still only part way through the editing process, but excited to announce Crafting Engineering Strategy, my upcoming book with O'Reilly on engineering strategy coming later this year (sometime Q4).
My bi-decade update on my desk setup 😂 Finally found a dock that lets me swap from windows desktop with dedicated GPU to OSX laptop with a single thunderbolt cable!
lethain.com/desk-setup-2...
This is a fascinating thread. And it has lots of my favorite thing: details.
I could give a shit about people who pounce online, flaming on and on about how wrong I am.
I love it when people reply with details about their very different experience.
The details give it heft. I can learn from them.
Coordination between company facilities teams and building’s janitor teams on how to clean rooms
Hah, I've heard of a few companies that did something similar! I don't think the _intent_ is wrong, but I do think it might be a lot more direct to educate the folks not sharing the space effectively.
Looking for startup to angel invest in that solves problem of chairs being routinely positioned to block video cameras in office meeting rooms.
(I've tried petitioning to remove chairs from rooms, but this innovation has met substantial resistance, so we need to innovate here.)
I respect you for that, but cannot join you lol
Each time I open LinkedIn I pray I don’t somehow accidentally click on the brain teaser game and become shamed forever off the internet
This time experimenting with MCP server for generating and running systems models ( github.com/lethain/syst... ). I think it's pretty wild how well this works, especially the tool to inject domain-specific language usage documentation into the context window.
Yeah… this is what I believe but if you told me I was wrong I wouldn’t argue too much 😂
