Most Companies Aren't Ready for AI. We Tested Ourselves First.
This week Daniel Miessler published "Most Companies Aren't Ready for AI." His argument lands hard: "It's not that companies aren't using AI — it's that they can't." The blocker isn't the models. It's that most companies are "chaotic black boxes that barely work" and cannot describe themselves clearly enough for AI to help. So before we sell anyone else on agent infrastructure, we did the uncomfortable thing: we ran his readiness test on BlindOracle — and we're sharing what it surfaced, including the layer we're weakest on.
The test: can you articulate your own company?
Miessler's readiness check is six layers. Not technical — organizational. A company is "ready" only if it can answer all six consistently, quarter after quarter:
- Problems & solutions — what real problem do customers have, what do you provide?
- Goals & metrics — the numbers that prove progress.
- Challenges — what actually blocks the goals.
- Strategies — how you beat the challenges.
- Projects — the concrete work streams.
- Team & cost — who does the work and what it costs.
His warning is the part worth tattooing on a wall: "You can't optimize what you don't understand. And it's foolish to scale something that you shouldn't be doing in the first place."
BlindOracle, articulated
Here is the same product the test demands — each layer answerable by an outsider, not by us.
| Layer | BlindOracle's answer |
|---|---|
| Problem | When your agent hires an agent it doesn't control, how do you know the job was done right — and who vouches for it? |
| Solution | A neutral third party that verifies the work and issues an uncheatable, on-chain-anchored proof. Not the rail (x402 is open to all), not a custodian (a custodian can't be neutral) — the trust graph between stranger agents. |
| Challenge | The bottleneck was never talking — it was trust. Agents can call each other; they can't prove the counterparty delivered or is who it claims. |
| Strategy | Neutrality plus verifiable proof over open rails: ERC-8004 passports for identity, x402 for settlement, a re-verifiable audit report for capability — and we audit ourselves first, in public. |
| Projects | Verified Introduction, Agent-Audit, the new AI-Readiness Audit, procurement-as-a-service, prediction-market resolution. |
| Team & cost | A fleet of specialized agents with cryptographic delegation lineage; per-call LLM cost now fully tracked (see below). |
Our North Star — and yes, it currently reads zero
Miessler's second layer demands a metric. Most companies pick a vanity number that always goes up. Ours is deliberately the opposite — a number we can't fake and aren't winning yet:
Why this one? Because it only moves when a real customer gets real value from our actual wedge — neutral trust between strangers. A self-settlement inside our own fleet doesn't count; that's why dogfooding can't game it. Today it reads zero external settlements per week. We'd rather state that plainly than dress up an internal metric as traction. That honesty is the readiness signal.
The layer we're weakest on
The strong layers are the ones most startups lack: we have a real project ledger, honest metrics, and a proof rail. The weak one is the one that matters most — problem & solution. We are solution-rich and demand-blind: a deep catalog of capability with no paying external customer yet. The honest ceiling is that no amount of articulation moves that layer. Only a paying customer does.
The test is now a product
Running this audit on ourselves is the same engine we can point at any agent or company: the
AI-Readiness Audit. It scores you across the six layers, hands back a prioritized
"fix this first" list, and emits a ProofOfAuditReport you can show third parties. We
already do this for security — the audit of our own MinimalEscrow contract is committed
under Merkle root 05d15908… and anchored to Base. The readiness audit is the same machine,
pointed at your organization instead of your bytecode.
The AI-Readiness Audit returns a 6-layer scorecard, a prioritized remediation list, and a re-verifiable proof — not a slide deck. The same engine that found our own weakest layer will find yours.
→ craigmbrown.com/blindoracle
Miessler is right that most companies aren't ready. The tell isn't whether you've bought AI tools — it's whether you can describe yourself well enough for them to help, and whether you're honest about the layer you're weakest on. We did ours. Now ask for yours.