Photos and videos
8/ We've felt these heebie-jeebies many times now. The team has remarkable unanimity on the cure:
When the shudder hits, do not search for the old tool.
Name the fear. Then go build the AI-era instrument that makes the fear obsolete.
#HeebieJeebies
1
4
9/ The teams that win the next decade won't be the ones who clung hardest to manual review.
They'll be the ones who felt the AI Vampire Heebie Jeebies — and used them as a map to the tools nobody's built yet.
More heebie-jeebies stories coming. 🧛 Stay tuned.
#AIVampire #HeebieJeebies
1
6/ The correct response isn't to reach backward. It is to invent the tool that doesn't exist yet.
In the AI era, your reviewer can't be 1-2 humans reading diffs. It has to be executable intent:
→ BDD — behavior-driven design testing
→ specs the machine can check faster than it writes
→ guardrails that run at AI speed, not human speed
1
6
7/ This is the part where the BC-era engineer says "but tests aren't the same as a careful human read."
Correct. They're better — because they don't get tired, don't rubber-stamp at 5pm, and scale to a codebase no human can hold in their head anymore.
The careful human read doesn't scale. Behavior does.
1
2
5/ The heebie-jeebies are a signal, not an instruction.
The signal: "our quality net has a hole."
The instruction your lizard brain hears: "go find the old net."
The old net was sized for a world where a human wrote every line. That world is gone.
#HeebieJeebies #AIVampire
1
1
4/ The math gives it away. A 3-engineer team shipping at AI velocity cannot eyeball every LOC. The moment you mandate it, one of two things happens:
→ you lie about doing it, or
→ you slow down to the speed of the thing you're trying to outrun.
Both are worse than the bug.
1
3
2/ At @TheSageOx, we call it the AI Vampire Heebie-Jeebies. #AIVampire #HeebieJeebies
It's the involuntary shudder an AI-native team gets the moment the stakes go up — and the body reaches back for the old tool it trusts.
For this engineer, the old tool is the manual human review.
1
3
3/ Here's the thing nobody wants to say out loud:
The shudder is real. The instinct it triggers is wrong.
"Put 1-2 humans on every line" is not a safety strategy. It's a séance — summoning the dead workflow of the BC era (Before-Compute) and asking it to protect a world it was never built for.
1
1
First of a series of stories on AI Vampire Heebie-Jeebies: #AIVampire #HeebieJeebies
1/ An engineer on an AI-native team sent me a message last week:
"We're starting to hit the point where operational bugs matter. We've been loosey-goosey with reviews for speed. I think every line that lands in production should be read by 1-2 humans."
This feeling has a name. 🧛🧵
1
5
At @TheSageOx we firmly believe that innovation and development has been and will continue to remain a team sport. The state of flow and joy which allows fearless innovation can only be attained if all the members of the team (humans and agents) are grounded in the exact same context using the hive-mind in real time.
5
Jun 12
We build at @TheSageOx with OxDot by our side. listening to everything we say. all day.
Most work conversations are now being recorded by default. You should probably assume that everything you say at work is getting recorded from here on out.
What’s emerging is a new category of enterprise software, organized around voice instead of text.
The system of record today is structured data: CRM entries, tickets, docs. But the highest-value context lives in conversation: the nuance on a customer call, the real argument in a product review, the offhand comment in a leadership meeting that quietly changes the roadmap.
LLMs are uniquely good at taking that unstructured voice data and making it structured, searchable, and queryable. That’s a large enterprise opportunity, and we’re still early in understanding what the software layer looks like and who owns it.
a16z GP David Haber on what AI recording means for the future of work: a16z.news/p/everything-is-re…
1
8
Jun 9
🧵Lean into your heeby-jeebies: HF Model Hub >> AWS S3
Hyperscalers ripe for disruption from ML specific cloud vendors like CoreWeave. Too fat, too slow for the current rate of innovations.
A story from my @huggingface days🧵
5
49
Jun 9
3/ The ostensible reason is separation of concerns: compute and storage should be kept separate.
The real reason, imo, as always is Conway’s Law: “Organizations which design systems are constrained to produce designs which are copies of the communication structures of these organizations”
1
16
Jun 9
Jun 9
Super excited to announce that @arcee_ai is the first major American AI lab to replace AWS S3 with Hugging Face for ALL their models and datasets, public AND private 🔥🔥🔥
Multi-million $ partnership to support American open-source AI, let’s go!
13
Jun 9
5/
6/2/26: @huggingface announces that Storage Buckets product is in line to hit an _exabyte_ by EOY
1
8
Jun 9
4
Jun 9
2/
4/13/25: Ajit @huggingface pitched AWS S3 for AWS native support for Content Defined Dedup.
Their reaction: it gave them ‘heeby-jeebies’ - couldn’t imagine AWS S3 supporting content defined chunking.
15