I help enterprises understand and use artificial intelligence. Leveraging my 25 years of enterprise software experience in emerging technology to drive results.
I've replaced about $1,000 a month in software subscriptions with tools I wrote myself.
Getting them ready to release as open source is taking longer than writing them did.
What's the first subscription you'd cancel if you could build it?
I use Anthropic's tools, and I have the token bill to prove it. I didn't start out confident in the company.
Reading the founders' earlier research changed how I think about their incentives. I think Anthropic could be the most important company of our time.
Before you test an AI model on a real job, write the right answers down first.
Renewal dates in messy vendor emails make a good test. A plausible wrong date is easy to accept when you haven't decided what correct looks like.
What's your go-to test job?
@steipete I feel your pain. Everyone assumes you're running agents on your everyday work laptop with your bank app open.
Let me choose to turn the alerts off. I hate when someone tries to protect me from myself.
Your next GPU order isn't in danger, so procurement will call this week handled.
It isn't. Move your top three AI suppliers onto the security risk register, with an owner who isn't in procurement. The exposure sits upstream: export rules, allocation, and the people at the top.
@NicolasBuilds@rauchg I don't think so I think that @letsencrypt solved this - https://t.co/HXHdL8F5Vd - I think most of the hosting providers including @vercel does this case in point - https://t.co/7Q1NGuY8WU - Though a couple of years ago I'd have agreed with you.
@rauchg ๐ฏ I have replaced about $1000/month in software subscriptions. Cleaning them up for release as open source when I have time, that took longer than it did to write the software. It's going to be wild next couple of years.
This is the problem I am seeing everywhere. We are moving from human-speed development to machine speed. My heart aches watching code get written in no time, only for my CI/CD loop to choke on PRs (and not just because of AI slop). I have no answer, but hopefully someday Anthropic and the rest of the industry will solve the problem. I also think developers will have superpowers if they can solve these problems (and they'll want to check out @poteto to see how she's doing it).
software engineers are going to have to tackle a new set of problems to unlock powerful multiagent systems to solve the world's hardest problems - right now is one of the most interesting moments in history to be an engineer. https://t.co/YsQ0xta4KR
OpenAI's October news reads like one plan: get AI working inside the apps your business already runs.
Some of it you can buy today with Oracle credits. The Oracle workspace is still being built.
I'd only put the first part on a delivery schedule. Where's your line?
Everyone's arguing Fable 5.2 vs Opus 5.5 on price per token.
Wrong number. Price the finished job: retries, fixes, and the person who checks it.
On my merges, the pricier model won. Has anyone run both on the same real task?
Near my Claude Max quota, I switched merges from Sonnet to OpenAI Dot with GPT-6 Astra.
Merges went from 0.83 an hour to 1.47. About 75% faster. The pricier model per token was the better value for the job.
Anyone enforcing a full SDLC on vibecoding seeing the same?
Gemini 4 Hydrogen is $1.50 per million input tokens. Cheap.
Cheap tokens don't promise a cheaper finished task. Count the retries and the person who fixes the output.
That's the bill I watch.
I had Claude run Jensen Huang's 92-minute GTC keynote through a free NVIDIA Nemotron model. Notes came back in 20 seconds. 12 of 14 details right.
One miss: nearly $10 billion turned into euros. Right number, wrong currency. Search the transcript before you repeat a number.
I spent a good part of my career in open source. The lesson from xz was that the code was fine right up until the person was compromised.
NVIDIA taught it again this week. I'm looking a lot harder at the people behind my AI suppliers.