David Chevenement

Tech entrepreneur, AI addict and curious about anything: science, space, finance, code, design, cars. CEO of @beesapps and creator of Beesy, AI for managers.

[David Chevenement @dchevenement]

@petergyang Hey Peter you don t need to. I’m using the 2 mouse thimb buttons and hooked them to the MIC and RETURN. Works like a charm ans no need to move my hands to dictate

[David Chevenement @dchevenement]

@mattpocockuk @weebhek Still amazed some people don t understand going over 150k context is making the model dumb !
Yet so easy to fall into this trap by keep pushing it further

[David Chevenement @dchevenement]

@mattshumer_ Sorry for you Matt.
I keep a 2TB external drive close by with timesystem on it. I plug it every couple of weeks. I did this in case my mac was stolen or lost while traveling. Most of my data is in sync on some cloud service but not all.

[David Chevenement @dchevenement]

@jonathan_wilke Because the ultimate interface is voice and you don’t need a ui for that. That’s how i interact with my agents since 6 months and this has been a game changer. The least is in the way the better, the agent will figure out how to do things without a UI in the middle

[David Chevenement @dchevenement]

@kr0der @jarredsumner Yeah they can support both agents.md and claude.md so no there is no smart reason to do this

[David Chevenement @dchevenement]

@paulg @synopsi LLMs definitely remind me of AIVAS in the Pern series of books from McAffrey

[David Chevenement @dchevenement]

@revolaition @lian75864 @petergyang @agents @AnthropicAI Just make a symbolic link to agents.md and you re set
Skills are a different issue though …

[David Chevenement @dchevenement]

@emollick Definitely a fake story
It would require 50.000 people at $10.000 each, not counting tiering and rate limiting from the providers themselves. We use AI extensively and we don’t reach this number on average

[David Chevenement @dchevenement]

@dangreenheck Hey Dan
This looks stunning to say the least. Are you planning on building a Pirate game out of it ? 👍

[David Chevenement @dchevenement]

@iamsupersocks Les LLM sont des machines a deplacer du PIB donc Mistral c’est la clé pour eviter d’en transferer une partie aux actuers americains et chinois.
C est effectivement un enjeu national (voir europeen) pour notre modele de societe demain

[David Chevenement @dchevenement]

@paulg @simonw Interesting take.
I have a Chief of staff agent which is writing some of my emails based on my dictation for the content but is doing the actual writing. This + a personal life file makes it exactly as i would write it myself 95% of the time.
Does this count ?

[David Chevenement @dchevenement]

@NicholasDClay @theo @johnsontrades_ $1.94 and $2.22 is the same price no ?

[David Chevenement @dchevenement]

@NicholasDClay @theo @johnsontrades_ Not at all. It costs the same on average as GPT 5.5 Medium. Hence Theo’s feedback.
We all expect a flash model to be around 1 Oom cheaper to consider the hassle of using it in our workflows. Otherwise what’s the point ?

[David Chevenement @dchevenement]

@mattpocockuk I’m fond of short skills calling scripts for deterministic actions, saving tokens and improving reliability. Your take ?

[David Chevenement @dchevenement]

@fchollet Is this based on actual data Francois ? On pur side we’ve seen huge Productivity gain for us with capabilities far outside of our reach before.
And we don’t measure in code produced but features delivered

[David Chevenement @dchevenement]

@theo CLI forces you to think Agentic first and to drop directly interacting with the system -> setting up your agentic and learning loop is mandatory. From there things only get better and better

[David Chevenement @dchevenement]

@bcherny @Taniyatweets_ It is indeed crazy. Used 2 accounts at 5x for a few prompts building some competition analysis. Explore agents are using sonnet. Both ended up at max 5hr window in 17 to 20 mn. It wasn’t happening before.
I had one 1x account, used it and it maxed out in 4 mn. All using opus 4.6

[David Chevenement @dchevenement]

@Trigun420 @bindureddy I was about to say that we’llbelieve it when we see it 😂

[David Chevenement @dchevenement]

@thsottiaux Hey Tibo
Great job congrats to the team.
You should build an agentic what’s new. Based on our existing pattern and workflows, would explain how we could enhance them with X or Y new feat.
Hardest part for us now is keeping up with new feats and ways of doing thing

[David Chevenement @dchevenement]

@thdxr Selfish people destroy everything. But it’s hard not to be selfish