I wrote a thing on my company's engineering blog about building an agentic system and specifically how I think about Trust. This really represents the v0 model for the system and we definitely need more optimizations/improvements. It's a solid start though. builders.cortex.io/blog/trustwo...
Building trustworthy agents for Operational Excellence reviews
How and why we built an agent to drive an engineering process
builders.cortex.io
I actually really admire Anthropic overall but I'm deeply concerned about sovereignty when running a frontier model in the US. This whole Fable drama is the realization of the risks that I and a lot of other people have expressed over market consolidation on US AI labs. Gonna get worse I think.
Man... The USG can be kinda dumb sometimes. It's not like we haven't seen a clear path towards progressively stronger AI models for the last 8+ years. It's moving fast... but nationalistic reactions are dumb.
Reminded today that duckdb is great when you need some random interconnect between random DBs on a bastion.
On my third side project in as many months and I definitely feel like it's a pretty cool time to be a builder. Undoubtedly pros and cons of AI, but it's made me more ambitious about silly projects and that is just fun.
Vendor lock in is obviously going to be a thing in agentic coding and systems. You can have your own harness or adapter layer but it's hard to say no to a fixed cost subscription. CostMaxing is gonna replace TokenMaxing
Claude is a stubborn mf sometimes.
rmoff 🏃♂️🫖🥓@rmoff.net · 4mo agoSometimes Claude has no sense of humour.
I really do wonder about the future of engineering teams and how dramatically we will see a scale down in their size. A lot of the friction we have in place right now is generally because scaling the number of _people_ working on a project means you have to create human-friendly guard rails.
Not to pile on, but the decline of GitHub is really sad to me. It's not like there was no more room to innovate in the space? Such a great brand completely run into the ground by a corporate overlord.
So... I think a lot of our assumptions about good software engineering practices are flawed. They are predicated on humans and humans ain't the only ones building software anymore. There's some things that probably DON'T matter anymore and some new things that our current practices can't handle.
A picture is worth 99.9 words. How's that @github.com migration from bare metal to Azure going? via Alex on Mastodon techhub.social/@ironicbadge...
I have a line in my global agent instructions about no random python scripts. AI usually follows the guidance, but tries to sneak them in occasionally just to keep me on my toes.
Still worried that the benefits of AI are forever going to be locked behind a bigcorp paywall. I thought it was wild that someone wanted $200 per month for AI. Kinda feels like we aren't even close to testing the market limits for the "best".
I've read some really bad takes from some well known engineers about AI agents like Claude Code. You are lucky if you've always had 10x roles with companies that had deep pockets to hire top talent. For the rest of us, $200 / month gives you crazy leverage to be more ambitious in your work.
It's a very exciting and scary time in software engineering. I simultaneously have a great sense of optimism and an overwhelming sense of terror at how much leverage an individual engineer has. The coin is spinning and no one REALLY knows if AI is going to end up good or bad.
I don't really dabble in psychology much anymore, but every once in a while I run into a phenomenon that reminds me why I was fascinated by the field. I just had a pretty great insight into a complex graph traversal problem that fell into place while I was brushing my teeth. Gotta love the aha.
One of my coworkers just got back from three months of parental leave. Bringing him up to speed on all the tooling changes with AI is kinda shocking. Not that it was bad three months ago but the capabilities REALLY have grown fast. I feel like a frog boiling in a pot of water.
The more time I spend in software the more I kinda wish I had a proper CS academic background.
I'm jealous of all the folks who get to go to kubecon this week. Have fun yall
It's been said before but one of the best features of LLMs are that they are more than willing to try things. It lowers the required inertia to try something novel.
Today a lot of folks found out that even though they don't personally host their stuff in AWS they are indeed indirectly an Amazon customer
I've always been of the opinion that ORMs abstract away the data model in a way that leads to short term gains and, inevitably, long term pain. Don't avoid thinking about your data model just because it takes extra work.
It low-key drives me nuts when people suggest using an LLM for something that good tools already solve well. Like linters. You don't need a language model to lint a million line codebase. You need a linter.
Momentum is absolutely a thing in engineering. The best products I've ever worked on built up a head of steam and had feedback loops that kept them pointed in the right direction. The worst products got bogged down in minutiae or, even worse, moved fast but in the totally wrong direction.
Today I was doing some toil work fixing a bug and, since it was the end of the day, I decided to let the AI have a go at it. The results were legitimately surprising.
Today's thread is about something I'm rather experienced with... How to Get Fired With Both Grace and Aplomb
I take a great deal of pleasure from explaining how people can lie with data. This takes the cake pmc.ncbi.nlm.nih.gov/articles/PMC...
Cigarette smoking: an underused tool in high-performance endurance training
The review paper is a staple of medical literature and, when well executed by an expert in the field, can provide a summary of literature that generates useful recommendations and new conceptualizatio...
pmc.ncbi.nlm.nih.gov
Convinced that my job is actually to write less code and that this is a good thing.
Say what you will about technology conferences but I just saw Wyclef Jean open for The Killers and they both were awesome