Guillaume Lebedel

It's probably just because they're selling you inference and they don't give a shit if only one of these gives you ROI

Everyone selling you inference is telling you you should be running 50 agents in parallel all the time.

@Matthewagi good point, i'll remind him that he's under constant threat of brain surgery

@glebedel that's a more polite first step than brain surgery tbf

My chief of staff agent has been autonomously creating a bunch of task for me in Linear and I'm starting to hate him. Do I talk to HR about this?

All you need is (good) MCPs

which makes me think ours is way too complex too.

very bearish on complex web UIs right now 😬

(Available OOTB for any of our 300+ MCP servers)

Our 109M param model - purpose-built for best accuracy and performance of Tool Search/Deferred lazy tool loading is #1 on ToolRet-full and a bunch of related benchmarks. Beats Anthropic Tool Use by 81%. Beats Qwen3-Embedding (8B). Beats NV-Embed (7B). Specific models for specific problems are worth it. @WillLeeney wrote a blog on how this model was built: stackone.com/blog/autoresea… "I do not have a tool to do X" is a surprisingly common problem when using any Agents. - whether it's because it truly doesn't have the tool, or because it has it but tool search of default harness sucks (eg. the one from @AnthropicAI's). @StackOneHQ solves both.

@EricNewcomer Funny what he took out from that was "should have asked another LLM" instead of...don't blindly trust LLM output and DYOR.

Our AI slop information ecosystem

The perfect modal Spot what's wrong.

@thegenioo Especially considering they’re - gonna make more on people that don’t move and deal with api usage - can repurpose compute for people that are willing to pay more but want a more stable product (enterprise) Sucks but it’ll take a lot more than this

@thegenioo 0.001% Of Anthropic’s revenue They’ll probably be alright

went through this Claude Sub cancellation thread from Theo 500+ replies, ~70% actual cancellations = 350 people gone (can actually be higher than this) rough math (assumptions): - 210 Pro @ $20 = $4,200/mo - 84 Max $100 = $8,400/mo - 56 Max $200 = $11,200/mo $23,800/month. $285K/year. From one tweet. and this is just the people who replied 💀

@boristane I even logged in via the AWS console mobile app just yesterday (I do not recommend)

@boristane Multiple times a day

trying to settle an argument how often do you open the dashboards of aws / cloudflare / vercel / fly / digital ocean / railway / render / <insert-provider-here> ?

Netflix has been super quiet in LLM land

@onerinas @linear (Turbopuffer is goated don't give me wrong, I'd have just assumed that they'd use elastic search for things beyond simple text search)

@onerinas @linear I'd be surprised. if they could replace what they originally added Elastic for with Turbopuffer. very interesting

Did ⁦@linear⁩ just replace Elasticsearch with Planetscale Metal?

@DavidKPiano The harness is the product.

Why does Claude not want us to use 3rd-party harnesses? Are they afraid we'll use fewer tokens than usual?