Philosophy

AI is leverage,
not a cheat code.

We’re human-first by design. We use AI to make people sharper, not to replace the practice that makes them good. Here’s the argument — and what it means for how we build.

The atrophy argument

Skills you don’t practice, you don’t have.

This isn’t new. Calculators didn’t kill arithmetic — but ghost-writing kills writers. GPS doesn’t kill drivers — but never reading a map atrophies your sense of direction. The difference is whether the tool compresses your reps or replaces them.
The replacement path

Hand the model the question. Paste the answer. Ship it. Six months in, the model has done all the thinking and you’ve done none. The first hard problem reveals you. So does the second.

The leverage path

Use the model to compress drafts, surface options, run scenarios — but make the call yourself. Six months in, you’ve done thousands of judgment reps with infinite raw material. You become the operator the model wishes it could replace.

The atrophy argument is why we don’t treat AI as a shortcut. We treat it as a forcing function — every cert path block is designed so the model accelerates the work but the human still does the judgment. That’s the reps that compound.

Human-first by design

The human stays in the loop. Always.

Every layer of our system has a place where a human has to think, decide, or sign off. Not because we’re anti-AI — because we’ve seen what happens when there’s no human.
01
The candidate is verified live

The Live Skill Sim is not a chatbot conversation. It’s a recorded voice interview where you, the human, have to actually think and respond in real time. The Six Cyborg Traits get scored from the recording — Override Instinct and Honest Uncertainty are the ones AI ventriloquism can’t fake.

02
Cert work is real work

Tier 2 deliverables aren’t multiple choice. They’re sandbox campaigns, copywriting, code, ops simulations — graded by AI rubric AND human review. We don’t pretend the model is the judge.

03
Decisions are explainable

When a placement happens, every input is on file: trait composite, sim recording, cert artifacts, authenticity signals. A human can defend the decision because a human made it.

Compute with intent

We use AI when it earns its keep.

The default in our industry is to run the biggest model for everything because someone else pays the bill. We don’t. Compute has a footprint — financial, environmental, ethical — and we account for it.
01
Right-sized models

We run Vertex 2.5 Flash for screening, Vertex Live for the sim, and the bigger models only where the marginal accuracy actually moves a placement decision. No frontier-model theatre.

02
Footprint-aware

A Tier 1 assessment on Flash costs less compute than streaming five minutes of HD video. We design new features against a stated compute budget per cohort — financial, environmental, and ethical considered together.

03
When NOT to use AI

Cert grading uses AI for rubric scoring AND human review. Final placement decisions go through a human. Compliance writing is human-drafted. Some loops are better with a person.

04
Local where local works

The architectural bias is local-first where the quality is acceptable — on-device transcription, cached embeddings for repeat queries. We don’t round-trip to the cloud for what a laptop can answer well enough.

“The opposite of AI maximalism isn’t AI rejection. It’s AI discipline. Most of what people use frontier models for could be done by a 7B model running on a laptop, and they’d get there faster, cheaper, and with less environmental cost. The cyborg argument isn’t use-more-AI. It’s use-better-AI, and use-it-where-it-actually-helps.”

The cyborg merger

Why the merger is the future
and not inherently bad.

Tools have always extended humans. The cyborg framing is just an honest name for what’s already happening — and a deliberate posture about how to do it well.
Tools extending humans isn’t new

Glasses, paper, the printing press, the calculator, the spreadsheet, the search engine — every one of them was called a cheat in its day. Each one made some skill obsolete and made another skill more valuable. The cyborg argument is the same argument, applied to the next tool.

The merger is unavoidable. The shape of it isn’t.

Whether AI makes the next decade better or worse depends almost entirely on whether the human stays in the loop and whether the tool is used to compress reps or to replace them. We’re betting on the version where the human stays sharp.

Specific knowledge still wins

Naval’s framing: specific knowledge is what society can’t train you on. The model can. So the moat shifts to: judgment, taste, context, the work you’ve actually shipped, the loops you’ve actually closed. Those compound. We’re built around them.

The cyborg is just a person who knows what they’re doing

No augmentation hardware required. The cyborg is the operator who uses the leverage and keeps the judgment. Everything we do is designed to produce more of them.

Use AI to get better at being human.

That’s the whole posture. If it lands for you, the next step is on the other side of one of these buttons.