Rahul

CAIO @codos_ai

Prev: Head of AI @lifiprotocol / @everclearorg / @Tesla / @NASA. 3x founding CTO.

[Dima @dimakhanarin]

Current situation in the "services-as-software" market.

[Rahul @RHLSTHRM]

@dimakhanarin Banger

[Rahul @RHLSTHRM]

@KSimback @JettMonroe0x Doesn’t more open source benefit their sales? If the new CEO is revenue minded they will stick with that.

[Rahul @RHLSTHRM]

The same agent setup gets accepted on the first try about 99.6% of the time on our people and HR jobs, and about 83% on deals. The gap isn't the model, it's the shape of the job, because a single request with a clear answer is basically solved while a recurring sync against messy data keeps finding new ways to drift.

If an agent job keeps missing, don't reach for a better model first, split it into smaller asks that each have one clear right answer.

[Rahul @RHLSTHRM]

I've gotten addicted to Grok Bot for personal use.

Spinning up bot teams and watching them communicate is pretty addictive, and I offload more and more stuff to them. This is the main agent I've stuck with after trying many commercial and open-source offerings, and building my own systems.

The crazy part is that no one rates the Grok models highly at all, Grok isn't even considered "frontier" at this point. Shows that the right UX shape and owning the user touchpoint is everything.

[Grok Bot @bot]

Introducing Grok Bot 1.0. AI teammates that work alongside your whole team.

Share Bots, collaborate with them in Slack and Teams, invite them to meetings, and watch them learn and become more helpful.

[Rahul @RHLSTHRM]

@KSimback Nice UI/UX is incredibly sticky. Employees demand this and will petition the company for it, even if it's more expensive/

[Rahul @RHLSTHRM]

@KSimback Would love to, but slack is just too sticky. You need the whole team to buy in to fully commit to using it, and with Slack Connect it’s ingrained with external partners too.

Slack is a perfect example of network effects that can’t be vibe-coded away.

[Rahul @RHLSTHRM]

@HochstatMichael We put risky steps behind approvals, and the part nobody warns you about is that people ignore them, since about 79% of our agents' approval requests in October never got an answer.

[Rahul @RHLSTHRM]

@kevinpiac What worked for us was splitting the work into small jobs and giving each one only the access it needs, service accounts too, so no single agent sees everything.

[Rahul @RHLSTHRM]

@afterxleep Proof covers the runs that finish, but about 93% of our agent misses since Nov 1 were runs that never finished at all, so we also watch for the one that just went quiet.

[Rahul @RHLSTHRM]

@goyalshaliniuk One I'd add is checking what it says when it stops, because one of our hiring agents burned about 95% of its budget and then signed off with "Stopped as requested" when nobody had asked it to stop.

[Rahul @RHLSTHRM]

One of our hiring agents used about 95% of its budget, then opened its answer with "Stopped as requested." Nobody asked it to stop, it was repeating our own wrap-up note back as if a person had pulled the plug, and our nightly QA run caught it before anyone went looking for who hit stop, which is how I learned the worst agent bug isn't a wrong answer but a confident one about something that never happened.

[Rahul @RHLSTHRM]

@UnpluggedTesla I’m in, check your email

[Rahul @RHLSTHRM]

@RShmider We do the same, clear the alert but keep the run record, because one timeout looks like noise and you only see the pattern if the history is still there.

[Rahul @RHLSTHRM]

@dimakhanarin The market is every company in the world.

TAM = Infinity

Every company can benefit by running agents.

[Dima @dimakhanarin]

I mapped 67 companies doing AI transformation.

Labs, Big-3, AI-consultants, Palantir alumni, YC startups, company brains and agent platforms.

7 things I learned:

[Rahul @RHLSTHRM]

In the past 2 weeks, about 93% of our agent misses were runs that never finished and about 7% were finished work a person had to fix, so I worry a lot less about checking outputs. The real concern is jobs that die quietly.

[Ethan Mollick @emollick]

I think there is way too much concern expressed about verification of AI output in everyday work due to a risk of factual errors.

It used to be true, but data on more recent models suggest they beat human experts, including PhD specialists, in finding facts and checking claims.

[Rahul @RHLSTHRM]

@mewwts Instant buy

[Rahul @RHLSTHRM]

@mewwts If you read owner testimonials, many who owned a CT said it was the best vehicle they had ever owned by far. I was really looking forward to new form factors of the “cyber” aesthetic, not just the cab.

[Rahul @RHLSTHRM]

Really sad, I was a couple months away from buying a Cybertruck. I’ve heard nothing but stellar reviews from owners.

[Elon Musk @elonmusk]

Tesla will be winding down production of the Cybertruck at Gigafactory Texas around the middle of next year to make room for increased Cybercab production