Links (6)

I am become Links, destroyer of productivity.

Disclaimer(s):

1. Just because I include a link does not mean I endorse it or am confident in its claims.

2. As an unrepentant capitalist, I use affiliate links for books.

Software Engineering & System Design

1. Everything I know about good system design: Sean Goedecke writes about what (he thinks) constitutes good system design. I share many of his intuitions, in particular on how good system design can be antimemetic.

2. How strong engineers break the rules and get away with it: Another post by Sean vocalizing some wisdom I've struggled to verbalize. In particular, "if you're breaking the rules then you need to be accountable for the outcome."

3. How I got promoted to staff engineer twice: Final post by Sean, which captures some of the tacit knowledge around getting promoted in large (think series B/C and above) engineering orgs. Description should not be taken as an endorsement by me or, I suspect, Sean.

4. Thoughts on the future of software development as a non-developer: Alex Telford on LLMs and software development. One of these cases where a semi-outsider is able to put things more lucidly than us insiders. Also, big congrats to Alex on the funding announcement! We at Stephen’s Exhaust Pipe have been proudly shilling Alex & Convoke since way before it was cool, and I have the receipts to prove it.

5. How I Use Claude Code to Ship Like a Team of Five by Kieran Klaassen: One of the many field reports I've read on using AI tools for programming recently. The question I keep asking, but not getting an answer to, is what does or will it look like for these tools to not just improve output but quality? This piece doesn't answer it either, but the anecdote about the queueing Ruby gem felt in the right direction.

   I asked Claude Code to look into the source code of the Ruby gem, a third-party code library we were using as part of the Cora app. It methodically walked through thousands of lines of someone else's code, step by step, and discovered something we'd have never found otherwise: The jobs were trying to line up under a different queue name in production, like packages being sent to the wrong warehouse. I might have been able to find it myself, after hours of digging through unfamiliar code. But Claude Code turned what could have been a daunting archaeological expedition into a guided tour, and we worked through the problem together. The AI did the research and dug through the source code, and we jointly came to the conclusion.

   As it turned out, there was no bug in our code. It was a mismatch between our development setup and the live website. But being able to work through that systematically was a breakthrough.

   This seems like as good a place as any to mention that I’ve recently been doing a lot of agentic coding myself on a personal project to try to build some intuition, and plan to write about it soon.

6. GPT-5: a small step for intelligence, a giant leap for normal people: If you read one thing to understand GPT-5’s implications for AI progress, I’d recommend this. There’s so much FUD out there these days, so I’m grateful to have someone like Peter, who neither exaggerates nor undersells. There is also substantial evidence that Peter is a substantially better-than-average, calibrated, accurate forecaster¹.

AI & Agents

7. Making of IAN v2 by Jan Hendrik Kirchner: Author, and Anthropic employee, writes about a V2 version of a personal AI he made. The original version was a fine-tune of GPT-J trained on his notes. As the few people who have tried this found, it was an interesting demo but lacked in coherence. This time around, he's using Claude Code on top of Obsidian (plus some additional stuff you can read about) and is finding it very useful. I continue to think this direction of AI tools is under-explored, probably because it doesn't appeal to enough consumers or enterprises so am very excited to see this experiment. I have just started playing around with something similar with my own Obsidian DB and I haven’t yet experienced that magic moment yet, but it has been fun and promising so far.

8. XBOW Unleashes GPT-5's Hidden Hacking Power, Doubling Performance: XBOW reports that GPT5 improved performance on their hacking benchmarks substantially despite OpenAI not seeing major improvements in their internal hacking benchmark. I think this is one of the first times I've seen a scaffolding make this big a difference (reliably). Perhaps this is the unhobbling we’ve been promised?

9. The Limit of Prediction is not Omniscience by Beren Millidge: Good points about the limits of perfect prediction on the datasets we have. A year or two "too late" but still instructive.

10. Agents and Lab Automation by Harry Rickerby: Useful snapshot of the challenges that come with using current AI to generate protocols and lab automation code. For now, UX and cognitive systems engineering remain underrated.

11. Billions Flow to New Hedge Funds Focused on AI-Related Bets by Peter Rudegeair: Continuing the trend of me using praise for others as a thinly veiled opportunity to brag about being early to recognizing promising things/people, go Leo!

Biotech & Healthcare

12. The new gene therapy playbook by Eryney Marrogi: My good friend writes what should become the telling of the miraculous story behind the n-of-1 gene editing therapy that recently seems to have cured KJ Muldoon’s carbamoyl-phosphate synthase 1 (CPS1) deficiency. As I said already, self-recommending.

13. Will all our drugs come from China? by Alex Telford: I finally read Alex’s thoughts on what’s going on with Chinese biotech. Characteristically lucid and informative.

14. The DNA protection company (Alan Tomusiak, Ep #4) by Abhishaike Mahajan: Good interview with the founder and CEO of Permanence Bio on a solid, under-explored approach to preventing cancer.

Strategy & Business

16. How to Run Smart Experiments When You Just Don't Know by Cedric Chin: (Paywalled) post by Cedric Chin on how to effectively FAFO (my term, not his). I suspect the ideal audience for this post are people like me who tend to over-systematize and need a way to justify and understand FAFO systematically. As an aside, I highly recommend Vaughn Tan's referenced book, The Uncertainty Mindset, a series of interwoven case studies on navigating uncertainty. The cases are all drawn from Vaughn's time spent embedding with top experimental chefs/in kitchens, which I found fascinating even as someone who’s not super into restaurants/food culture.

17. The Chaos-to-Conscientiousness Phase Transitions by Chris Barber: This is very good. I recently joked, “I know people who'd be better off working less hard and exploring more. I know others who'd be better off buckling down and just focusing on one damn thing for a bit. The former are mostly friends from college, the latter friends from the internet.” This post captures how this dichotomy interacts with company phase progression well. I also think the traits that work best are sectoral, and so it'll be interesting to see how the tech to bio migration, which I'm a part of, looks in hindsight.

18. How to hire smarter than the market: a toy model by Erik Bernhardsson: I reread Erik's old post on how to hire smarter than the market and liked it even more the second time around. It’s pretty wild how Erik wrote this post and then proceeded to start an awesome company that basically every engineer I know agrees has hired one of the most talented engineering teams in the startup world. Makes ya think!

19. Strategy & Non-Financial KPIs by Roger Martin: One of these topics where there’s a million unhelpful posts treating the 101-level stuff and basically none about the nuts & bolts that everyone who tries to do this for real runs into. While the post won’t be mind-blowing for people who have had to figure stuff out themselves, I wish I’d had it 10 years earlier, and so hopefully it saves someone the pain many of us had to go through.

Philosophy & Epistemology

20. On Pessimization: Richard Ngo writes about "pessimization", the phenomena of agents working against their own, or others', goals. Richard is one of the few contemporary successful torch carriers of coining new, useful (to me) concepts. This defined much of the blogosphere at one point but became harder as many of us grew wary of the difference between insight and insight porn. I suspect a key differentiator of Richard’s approach is that he’s willing to give real, often controversial examples, whereas others often stick to abstractions or hypotheticals to avoid angering people.

21. Which Ways of Knowing Actually Work? by Linch: Good taxonomy of ‘ways of knowing’. However, what about reasoning?!

Government & Industry

22. How to Be a Good Intelligence Analyst by Santi Ruiz: Fascinating nuts & bolts discussion of the intelligence community and its interactions with politicians. I was pleasantly surprised by how much of the interview focused on tacit knowledge and forecasting.

23. Building an American TSMC by Contrary Research: Long report on the potential ways to build an American TSMC. I admittedly have only read a small fraction of the full report but found it well-researched and novel.

24. Funding Open Source like public infrastructure by Dries Buytaert: One of my (boring) conspiracy theories is that there was a concerted effort to memory hole the fact that open source was originally intended to mean “free as in free speech, not free as in free beer”. Towards combating this wrong-headed idea, Dries has long argued that, in fact, open source can only be long term sustainable if there are viable means for people to get paid to work on it.
    
    Here, he argues that governments should not just use but support and fund open source software to promote its long term health and robustness. I knee jerk am more wary of relying on government as a primary open source funder and am more attracted to many of Vitalik ideas, but I also think it’s good Dries is making these arguments and buy that on the margin open source is a totally worthwhile thing for governments to fund. In particular, I think US research funding has historically under-funded/incentivized high quality open source work relative to papers.

    Fifteen years ago, I laid out a theory about the future of Open Source. In The Commercialization of a Volunteer-Driven Open Source Project, I argued that if Open Source was going to thrive, people had to get paid to work on it. At the time, the idea was controversial. Many feared money would corrupt the spirit of volunteerism and change the nature of Open Source contribution.

    In that same post, I actually went beyond discussing the case for commercial sponsorship and outlined a broader pattern I believed Open Source would follow. I suggested it would develop in three stages: (1) starting with volunteers, then (2) expanding to include commercial involvement and sponsorship, and finally (3) gaining government support.

1. Peter is a globally ranked top 20 forecaster across multiple platforms.