Chris Parsons
@chrismdp.com
I help leaders leverage AI in their organisations - see chrismdp.com đź’Ş also co-founder & CTO of cherrypick.co
I used to use Claude for every piece of work and hit the limits within hours. Now I script repeatable tasks so the agent can run them without reasoning them out again.
The most important thing about any system is figuring out how to record structured feedback from the people who understand it, so that an AI can continually improve the system. https://www.chrismdp.com/prompt-evals-are-useless/
Prompt Evals Alone Are Useless
Over the last couple of nights I have been building a text adventure RPG, mostly to see what Astra could do. I have been giving the AI an end of day prompt, ...
chrismdp.com
I cannot believe how much usage I'm getting from Opus 5.5 compared to Astra.
OpenAI has shelved GPT-6.1 Astra. Its safety lead says the model fell short on staying within its authorised scope and telling users what it had done.
I now do 95% of my work in Claude, WhatsApp and Telegram. https://www.chrismdp.com/claude-code-on-your-phone/
Run Claude Code From Your Phone
Your laptop is a terrible development environment. It sleeps when you close the lid, loses state when you reboot, and chains you to a single device. The solu...
chrismdp.com
Want to learn exactly how LLM work? I'd highly recommend Karpathy's videos building an LLM from scratch.
Some intel on where Jev might fail to work for you (and why that doesn't always matter):
Productive individuals do not make productive firms. AI rollout work has to become shared operating practice, or it stays as private cleverness.
I can't stand the coder gatekeeping I see everywhere! Engineering is no longer out of reach with AI.
all set up for my Design+AI workshop today! Teaching designers how to think like an engineer in their production projects - I’ll post a proper write up of session soon
Earlier this month I left the country for four days. My AI kept the business running while I was gone.
Trump is right: "artificial intelligence" is the wrong name. "Machine Intelligence" is much better than his options though.
It looks like a bet on a) ever increasing AI capability is possible given more compute and b) demand keeps up with the capability on offer within the investment cycle. Remains to be seen whether this is a good bet: seems high risk to me
Renaming Artificial Intelligence is definitely not the most important thing on the US president's agenda.
I've had great success with adding an automatic transcript check before adding the transcript to your personal file vault: given the context of the vault, it will often correct misspellings and prevent the proliferation of bad information.
Really impressive demo from Ethan Mollick of using Blender to create a storyboard and then upload that to a video generator - reminds me of my using Claude Design but on steroids https://www.chrismdp.com/software-factories-can-handle-ux/
AI progress still compounds. The chart is now current; the cautions about over 16 hours are on it too.
Trying lots of open models today, mostly because I've managed to hit limits on 2 OpenAI Max 20 and 1x Claude Max 20 accounts!
I spent an evening testing TypeSafe's Jev against the models I already pay for. 16,000 decisions across eight real tasks. Fast and cheap, and I am still not putting it near production.
Your prompt passed your evals, but your product still failed. The fault may not actually be the prompt.
My half-day workshop at design + AI in Brighton next week has sold out!
I'm no longer a coder, I'm a director of change. I rewrote my "coding with AI" article. It needed an update after only 5 months, because the job has now entirely changed.
Not sure Apple has quite got this right. Not sure we're ready for summarise-by-default, although it will be undeniably useful.
Agents keep asking the same questions until you tell them how you work.
