Jean-Baptiste Coger

Building @vybe_build: AI agents your whole team uses, in Slack. French founder in SF. Previously cofounded @platohq.

[Jean-Baptiste Coger @jb_coger]

@jans Congrats on the launch! Turning a 30-second voice note into tasks for the right agents is a neat idea.

[Jean-Baptiste Coger @jb_coger]

@DanielLenton1 @letsunifyai Congrats on the launch! Great to see someone actually test whether agents learn from experience.

[Jean-Baptiste Coger @jb_coger]

@sochak03 @ycombinator @STASISYCF26 @neil_gadia @kul @garrytan Congrats on the launch, Soham!

[Jean-Baptiste Coger @jb_coger]

@ReticularAI Congrats on the launch! Love that you published a public replication.

[Jean-Baptiste Coger @jb_coger]

@hatchet_dev Congrats on the v2 launch and the raise!

[Jean-Baptiste Coger @jb_coger]

@tryaircaps Congrats on the launch! Love that you open-sourced the benchmark.

[Gabriel Chua @gabrielchua]

had fun building this deck in ChatGPT Sites for a talk on dots today… or rather, working with my dot to build it 🙂

demoing dots is hard. A lot of the personal and work context is private, and you can’t exactly schedule a proactive heads-up for the middle of a live demo.

so I picked a few stories about how my dots have helped me. I basically asked my dot to recount what happened in our chats, then recreated those conversations on the slides one message at a time.

from finding me two last-minute seats on an earlier train, to chasing for various refunds, to spotting an invoice that was higher than the quote.

And on the point of proactivity, my dot gave me a heads-up just before the talk to keep a backup demo handy

[Jean-Baptiste Coger @jb_coger]

@hwchase17 At Vybe we treat it as both. Agents write what they learn into plain notes and skills that the team can read and edit, and every correction gets saved with the reason behind it. The hard part is still invalidation: last month's rule can be wrong after this month's product change.

[Jean-Baptiste Coger @jb_coger]

@gabrielchua This is so cool! Replay is what makes browser agents usable for real work: you can actually see what happened instead of hoping it went right. Can't wait to try it.

[Jean-Baptiste Coger @jb_coger]

@gregisenberg Same shift here. I walk home from work now and turn it into 30-minute working sessions with Codex over chat. Some of my best work of the day happens on that walk.

[Jean-Baptiste Coger @jb_coger]

@sydneyrunkle How do you set permissions for this today? Per card, per merchant, or a confirm step before anything gets charged?

[Jean-Baptiste Coger @jb_coger]

@alexeyeryshev @ycombinator Congrats on launching Munito Sandboxes!

[Jean-Baptiste Coger @jb_coger]

@celltypeinc @NVIDIAAI Congrats on launching BioSimBench!

[Jean-Baptiste Coger @jb_coger]

@hwchase17 Matches what we see. One trace failing is easy to catch. What's useful is grouping thousands of traces by use case and spotting the failure that keeps coming back. That's analytics, not alerting.

[Jean-Baptiste Coger @jb_coger]

@dexhorthy AI slop is delegated workload. The sender skips the work of choosing their words, and the reader pays for it.

[Jean-Baptiste Coger @jb_coger]

@Andrey__HQ Investors replying unprompted because they watched you ship every month beats any deck. Congrats on the October numbers.

[Jean-Baptiste Coger @jb_coger]

@gregisenberg Same thing in agents. Nobody wants "an AI agent." They want the Monday report they hate writing to already be done.

[Jean-Baptiste Coger @jb_coger]

@gabrielchua The interesting part is deciding what deserves to wake you up. Does dot have a clear boundary between 'keep going', 'leave a morning note' and 'stop until I approve'? That's the bit I'd want to understand before leaving it overnight.

[Jean-Baptiste Coger @jb_coger]

@bradneuberg The missing cost in the 15-agent setup is the human attention bill. I'd rather measure useful work per interruption than output per hour. What would you stop delegating even if the agents got twice as fast?

[Jean-Baptiste Coger @jb_coger]

RT @qhoang09: Today we’re launching Vybe to the world and announcing our $10M Seed round to make vibe-coding actually work inside companies…