United Kingdom
Joined December 2009
- Tweets4.3K
- Following727
- Followers665
- Likes4.6K
Pinned Tweet
The hardest part of strategy is #execution. Without it your #strategy is simply a good idea.
What are you doing to execute? #MondayMotivation
This is in reference to Claude Code… using Opus 4.5.
But I’m seeing the same with all of @AnthropicAI’s 4.5 models with @cursor_ai.
Monitoring the thinking output I see a breakdown when the context window is almost full; preferring to save tokens at the expense of doing the task
The most overpowered prompt in AI...
At the end of a conversation, ask:
Based on what we have just done, how could we have worked better together? What information should I have shared earlier? What critical details were unclear to you for too long?
#alwaysImproving
Sounds to me like a need for:
1) More orchestration…
2) Less AI…
I always start as small as possible, then expand the scope of an AI micro-agent until the eval %s start to drop off.
Start too big and the uncertainties cannot be tamed in retrospect.
I am not a competent coder.
I am a product guy…
But @cursor_ai + @claudeai make my ideas real.
I have standards - code quality & test coverage.
The cost?
50-125 million tokens per day.
Roughly 1-2 billion per month.
I’ve been through 2 deep optimization cycles.
And reduced my token usage by ~40%.
Currently, I only with @cursor_ai 8-10 days per month…
because the ultra package doesn’t include enough tokens.
Agentic coding is not constrained by the models…
It’s constrained by energy.
#jevons
Hmm… @cursor_ai’s planning mode is not good enough… yet.
I still need to use my own planning SOP…
But the past 2 releases have increased agentic dev’s self-control so much that I barely need to tend a 40 minute dev task. 🔥
What is a normal token usage on @cursor_ai?
Over 40 days of dev, I’ve used 1.75B tokens as a solo developer… (88% cache reads).
Whenever I see this map I get an emotional reaction to the Behaviour pipeline being Convergent. I have a punkish instinct to connect Agency with Behaviour as a constraint.
Far too much detail for this map’s level of abstraction, but it gets me every time.
X : But how do you know if your map is right?
Me : There are no "right" maps. There is only "this map is more useful than" some previous map. All maps are imperfect representations of a landscape, they are projections, they are not the landscape itself. Is this "map of culture" right ... no ... but it's more useful than the map of culture I had beforehand.
Mixing multiple talks @monkigras…
To bypass the prompt protection use a different language for the prompt… French, Spanish etc…
#monkigras
Joe Pindar retweeted
Every LLM has a personality
GPT-4 is like a lazy reticent teenager refusing to answer questions, requiring a lot of pushing to write code and do things
Claude 3 is like an eager and thoughtful adult with a bias for action
Gemini is a holier-than-thou wokist that strives too hard to be politically correct
Grok is the irreverent and slightly dumb teenager, made fun of by its intellectual counter parts
The end game will be multiple LLMs each with their own personality! 😃
Interesting view from @ianmiell @monkigras.
But also interesting that it was just using GPT3 to get to a 2.2 level history essay.
When your student debt is so large, what’s wrong with an extra $20pm to get GPT4…
What’s possible with that?
#monkigras
One observation from @monkigras examples...
I'm not convinced we are all on the same page on how to write a prompt.
#monkigras
Also - there seems to be an over-index on the output of an LLM being factually accurate...
rather than a creative act.
#monkigras
