La Batte

Founder @reflect_hr + Product / AI / Tech / Design

[La Batte @baptjan]

@tainmar Ouais la honte

[La Batte @baptjan]

Vous en êtes où niveau IA chez Reflect ?

On est à l'état de l'art. Tout va bien.

[La Batte @baptjan]

@orca_build We need webhook automations, not just cron.

My team wants to use orca to manage all our agents but because of this we are still using claude code and openhands…

[La Batte @baptjan]

@charles_maddock It removes all claude comments

[La Batte @baptjan]

found a comment in our company wiki this morning. posted at 3:12am. no author.

it's addressed to "the other agents"

we only have one agent

[La Batte @baptjan]

@abbeyjazzy 😅 like this
did the same joke for "ajean delevoye" yesterday

[La Batte @baptjan]

my ai agent and my cofounder's ai agent booked a 2h meeting this morning to settle our q4 roadmap.

we're both marked "optional"

they already reached consensus

[La Batte @baptjan]

@steipete @simonbs Do u know u can also do both: mini + tailscale for multi users

[La Batte @baptjan]

@joenrv I think you can now post images with gh cli. Also my agent is posting automated videos made with playwright instead of screenshots: this is 🔥

[Mnimiy @Mnilax]

it's f*cking insane

Jev sat between GPT and me, killing every draft that broke my rules before i saw it. good setup.

then Jev did not answer. the agent decided silence was safer and stopped sending me anything at all. took a second agent to unstick it.

your checker needs a branch for the moment it goes quiet:

> let it through tagged unverified and keep moving
> hand the risky span to the big model
> park it in a queue and retry in a minute
> wake a human when the action is irreversible

70-500 ms

one Jev call, by TypeSafe's own number. a half-second checker still takes down the whole pipeline when it has no default.

[La Batte @baptjan]

@vicgrss probably 88 IQ

[La Batte @baptjan]

applied to a silicon valley ai company this morning.

got a reply in one second: "No (0.88)"

they already plugged jev into their ats

[Addy Osmani @addyosmani]

"The most important skill in prompting is expertise in the domain you’re prompting for"

https://t.co/Dgme0HcL2z

Sometimes the human is the bottleneck. A good reminder LLMs are a force multiplier if you're skilled and to build this domain expertise if you don't have it yet.

[La Batte @baptjan]

Half of my messages to Claude are: "Take the PR feedback surgically and keep the commit history clean."

[La Batte @baptjan]

When investigating, my orchestrator agent biased every subagent it spawned. had me diagnose a regression that never existed.

As on code reviews, i now make it spawn them explicitely unbiased: minimal context and the question, nothing else. no hypothesis, no hint of what is expected.

I'm getting much better results with @claudeai

[La Batte @baptjan]

@Recursive_SI just put $400m of its $650m raise into aws compute. @RichardSocher said "it's less about headcount and more about agent count".

honestly the dream for #hranalytics. no headcount to reconcile, no leavers, no duplicates. just one line on an aws invoice (going up).

https://t.co/j9zByr6Xl2 via @TechCrunch

[La Batte @baptjan]

@ZidaneOfficial_ just got promoted 20 years after headbutting a coworker on live tv in front of a billion people. @FFF

Remember that next time HR tells you a slack message can end your career.

[La Batte @baptjan]

@OpenAI just launched ChatGPT Work. Agents now work inside your tools: @-mention an app, they pull the context.

Your email has a connector. Your CRM has a connector.

Your payroll? Nothing. So people export it to Excel and paste it in. Dirtiest door possible.

We built the clean one: @reflect_hr , the only HR data MCP. Read-only, permissions, logs. Live on ChatGPT, Claude, Dust, Copilot.

[La Batte @baptjan]

400+ connectors in @claudeai's marketplace.

CRMs, warehouses, design tools, code. Everything.

HR software? @Workday , @bamboohr , @Rippling : nowhere. The industry sitting on your salaries still doesn't speak AI.

As of this week there's one exception: Reflect. The only HR data MCP in there.

[La Batte @baptjan]

Your turnover is at 10%. Good news or bad news?

10% turnover rate, hires and exits combined, might just mean you're growing fast. Healthy rotation, nothing broken.

10% voluntary attrition, people leaving without you choosing it, is a completely different signal. Talent walking out the door you didn't see coming, for reasons you haven't diagnosed yet.

Same number, two opposite situations, two completely different problems to fix.

The metric you track is the problem you solve.

And that's not a detail, it's the whole game.
HR data is the most direct signal a company has on its own behavior: who stays, who leaves, why, when.

But a signal is only worth something if the definition behind it is airtight. An badly defined metric doesn't give you bad data. It gives you confident data pointing in the wrong direction, which is worse.

That's the wall every HR team hits when they try to operationalize their data. Not the model. The semantic layer underneath it.

At Reflect, turnover and involuntary attrition are separated explicitly. Documented formula, fixed perimeter, versioned over time. So when the number moves, you know if it's reality shifting.