Alex Balashov

"... we are trapped with a berserk brontosaurus peddling vehement ignorance." tinabrown.substack.com/p/trumps-wild-…

My presentation about the past and future of the American telephone network given on 6 Mar at @socallinuxexpo/@astricon 2026 in Pasadena, CA. I'm sure a recording will be forthcoming relatively soon, but in the meantime, here are my slides. scribd.com/document/10099…

"Foreign exporters absorb only about 4% of the tariff burden—the remaining 96% is passed through to US buyers. Using shipment-level data covering over 25 million transactions valued at nearly $4 trillion, we find near-complete pass-through of tariffs to US import prices." kielinstitut.de/publications/a…

"Gaslighting is just one of the 'therapy-speak' terms that couples therapists told me their clients are misusing, typically after seeing descriptions of the ideas on social media." theatlantic.com/family/2025/12…

@montef Ha! The funny thing is that I have -- I just wrote an agent that ties Claude and an MCP server together to do code review in a version control/CI-CD workflow. It is impressive and useful, but it stops quite a bit short of economically transformative...

@abalashov Oh. I see the problem. You just haven’t used Claud paired with a Grok MCP and the latest OpenAI pipeline models yet…… 🙃😉 In seriousness though, I think it’s going to be ugly when the AI bubble finally bursts. 😖

(1/10) The central tension of #AI-assisted programming seems, to me, to be: one cannot competently evaluate the code output of #LLMs, nor to competently steer them as an architect, without having banged one's head problem-solving in code "the hard way" for a few decades...

(10/10) market can remain irrational far longer than some of us can remain solvent, it is wise to ignore the FOMO and snake oil, stick it out, and hang onto common sense for dear life.

(9/10) chart an independent course on AI based on a blunt understanding of the bottom line and a sober, cold-eyed measure of the business results LLMs really achieve, or don't achieve for you. I'm not saying you shouldn't use it--that ship has sailed. However, and although...

(2/2) ... about productivity gains and embracing exponentials, disappearing for a few months on some vibe coding ayahuasca type vision quest. They emerge battered, bruised, a little wiser perhaps, and with a more ambivalent view of LLMs as an economically transformative force.

(1/2) A few times now, I've had the experience of highly technical friends discovering Claude Code for the first time, declaring that it's all over and the singularity has arrived, speaking euphorically--religiously, agitatedly, transcendently--...

"... new review of 15 years of research concludes that the evidence of its benefits is often weak or inconclusive, and that nearly 30 percent of medical cannabis patients meet criteria for cannabis use disorder." nytimes.com/2025/12/12/hea…

(3/3) The specificity and the details here are extremely important, and there's just no skating around it.

(2/3) LLMs are useful, but not quite in the ways most think. LLMs will be economically disruptive, but not in the ways or to the degree most people imagine. LLMs are transformative, but not in the areas most untutored commentary suggests and not to the degree prognosticated.

(1/3) The moment of #AI has exposed, more than anything else in recent memory (except perhaps the Covid pandemic), the limits of an economic and political culture built on broad pronouncements without a hint of nuance.