Whoever Defines the Work Primitive Wins.
Sourced from Nate B. Jones’s essay “Access Is Not Meaning”. As compute access commoditizes, the durable moat moves up the stack — to whoever defines typed, permissioned, reviewable work primitives that agents act on.
The Spectrum
Five tiers of agent-readiness, from least to most semantic:
Tier 1
Pure Access
Tier 2
Access + Inference
Tier 3
Partial Semantics
Tier 4
Rich Semantics
Tier 5
Platform-Grade
The Four Species
Every AI product, company, and operator falls into one of these. Knowing which one you are — and which one you’re building toward — is the strategy.
Skilled human + frontier model
Brilliant Operator
The operator is the moat. They wield agents the way a master craftsman wields tools — speed and judgment compound. Their leverage comes from taste, sequencing, and context, not from owning a platform.
- When this wins
- When the work is high-judgment, low-volume, and the operator’s name is on the output.
- Risks
- The operator is the bottleneck. Hire them, lose them, and the magic walks out the door.
- Examples
- Solo CEOs who run companies through Claude. Designers who ship in hours. The maestro pattern itself.
Owns the work primitive others depend on
Semantic Platform
Defines a typed, permissioned, reviewable primitive that other agents (and other companies) build on. Stripe’s payment token. Plaid’s account link. The primitive becomes the ontology.
- When this wins
- When you can codify a unit of work so well that downstream agents would rather call your API than reinvent it.
- Risks
- Hard to bootstrap. Requires the work primitive to be both general enough to compose and specific enough to be useful.
- Examples
- Stripe payment intents. Plaid item tokens. Anthropic’s tool-use schema. The mpowerio substrate’s tenant-scoped audit log.
Source of truth that exposes itself cleanly to agents
Agent-Ready System of Record
Where the data actually lives. The truth about a guest, a trip, a vendor, a job. Exposes typed APIs, audit history, RLS-enforced reads, and explicit write surfaces — so agents can read and act without breaking invariants.
- When this wins
- When your customers' operational data is locked in spreadsheets or legacy CRMs that hostile-by-default vendors won’t open.
- Risks
- Data migration is brutal. Trust takes time. Compliance, retention, and recovery have to be airtight.
- Examples
- The mpowerio substrate. Supabase + RLS + audit_log. The boutique luxury travel agency’s 658 guests + 832 vendors after the 2-hour migration.
Owns the data, refuses to expose it
Agent-Hostile Incumbent
The incumbent that monetizes user lock-in. Won’t open APIs. Throttles bulk export. Slaps lawyers on scrapers. Their moat is friction. Agents make this stance increasingly untenable — but it can persist for years.
- When this wins
- Never — this is what we displace.
- Risks
- Vulnerable to a Semantic Platform that customers prefer. Vulnerable to regulation. Vulnerable to a competitor that opens up.
- Examples
- The CRM your client begged us to migrate them off of. The point-solution that emails CSVs but won’t expose webhooks.
mpowerio Is the Semantic Platform for Small Business.
We don’t just give you AI — we give you the semantic moat. RLS = permission encoding. ENUM stages = action vocabulary. Audit log = validation history. Tenant-scoped data = blast-radius scoping. The brain layer = your owned memory.
Open-weight Llamas running locally on commodity hardware don’t disrupt this — they amplify it. Access becomes free and sovereign; meaning (the substrate) becomes the durable layer.
“We don’t manipulate. We orchestrate.”