---
type: digest
format: essay
source_note: URL not provided — captured directly from pasted essay text (likely Greg Isenberg Substack / blog / LinkedIn long-form)
author: Greg Isenberg
captured: 2026-05-11
posted: 2026-05-11
tags:
  - ai-native
  - agentic-ai
  - enterprise-ai
  - workflow-redesign
  - vertical-ai
  - services-as-software
  - revenue-per-employee
  - small-teams
  - operating-model
  - forward-deployed-engineer
related:
  - "[[05 People/Greg Isenberg]]"
  - "[[08 Reference/2026-04-24 Greg Isenberg on AI Repricing and Startup Surface Area - Digest]]"
  - "[[00 Raw/X Posts/2026-05-11 Aaron Levie - Professional Services and FDEs for Agents]]"
  - "[[00 Raw/X Posts/2026-05-11 Aaron Levie - AI Automation Engineers at Box]]"
  - "[[08 Reference/2026-05-11 OpenAI - Launches the Deployment Company - Digest]]"
  - "[[09 Implications/Migrate Workflows Not Seats]]"
  - "[[09 Implications/Pair Governed Runtimes With Workflow Owners]]"
  - "[[08 Reference/Agent-Native Vertical Software - Headless SaaS as Backend]]"
  - "[[03 Research/AI Infrastructure and Agentic AI - The Control Surface Thesis]]"
---

# Greg Isenberg — How to Become AI Native

**Author:** Greg Isenberg
**Captured:** 2026-05-11
**Source URL:** not provided at capture

## Core thesis

An "AI-native" company is **not** a company that uses AI — it is a company that has been **rebuilt so AI can operate inside it**: structured, documented, permissioned, and instrumented in a way that agents can understand. The company has made itself **legible to machines**. Most companies today are barely legible even to their own employees, which is why their AI investment can't transform them. The opportunity isn't building AI products; it's redesigning the operating model of the company itself so agents become part of the operating fabric, not a bolt-on.

## Key takeaways

- **AI-assisted vs. AI-native** is the load-bearing distinction. AI-assisted companies add AI at the **edges** ("where can we save time?"). AI-native companies redesign the **center** ("how should this workflow exist if agents do the first 80%?"). The second question rewrites the org.
- **"Maybe 1,000 truly AI-native companies on earth"** doing $5M+ ARR. Could be 500. Could be 2,000. The point: despite the noise, the field is structurally empty. "Noise is not maturity."
- **Why most companies aren't legible to machines.** Truth is scattered across CRMs, Slack threads, personal inboxes, Notion docs nobody trusts, spreadsheets named "Final_v7_NEW", tribal knowledge ("talk to Sarah"), and undocumented exceptions. "AI cannot run on vibes."
- **Reframe: a lot of "work" is organizational friction wearing a fake mustache.** Humans currently spend their day moving information between tools, remembering process, searching folders, classifying requests, rewriting emails, chasing approvals. AI-native companies strip that out and keep humans for taste, trust, judgment, relationships, and exceptions.
- **New management model, not a productivity improvement.** For 100 years scaling meant hire more, add managers, buy software, invent coordination process. AI-native companies scale as **small teams running fleets of specialized agents**. A 12-person co. does what used to take 80; a 40-person co. competes with a 400-person incumbent.
- **Revenue per employee becomes the dominant signal.** Also: gross margin, resolution time, conversion rate, error rate, customer satisfaction, sales velocity, onboarding time, renewal rate. AI-native shows up in the numbers, not in branding.
- **Incumbents can't retrofit easily.** "An old company cannot become AI-native by announcing an AI initiative. That is like trying to turn a cruise ship into a speedboat by buying a new steering wheel." The blocker isn't model access — it's process debt, conflicting policies, turf, headcount-shaped workflows, and an OS that assumes humans are the default processors of information.
- **The opportunity hides in boring categories.** "Weird little money machines" — narrow markets, proprietary workflows, high automation, boring category, beautiful margins. Specific candidates named: AI-native agencies, brokerages, law-adjacent services, accounting firms, compliance shops, healthcare admin, real estate ops, education services, logistics coordinators, BPOs that don't look like BPOs.
- **They will look like services businesses with software margins hiding inside.** Investors and competitors will misclassify them. That's an advantage.

## The five-step playbook (verbatim structure)

1. **Pick a narrow workflow with obvious economic value.** Support resolution, outbound prospecting, onboarding, claims intake, document review, renewal management, reporting. High volume, existing rules, currently human-heavy coordination.
2. **Map the workflow like a machine.** Triggers, required data, decisions, reversibility, approval points, success criteria, error modes, tacit human knowledge.
3. **Structure the knowledge.** Write the policy. Make pricing rules explicit. Clean the customer object. Create examples. Define tone. "This is where most teams quit, because it feels like documentation. It is not documentation. It is infrastructure."
4. **Put agents in the workflow with boundaries.** Let them draft, classify, recommend, enrich, summarize, prepare. Action authority only where risk is understood. Approval gates for judgment. Log everything. Track quality. Improve the system.
5. **Measure the business impact.** Not vanity "hours saved" — real KPIs: resolution time, conversion, gross margin, revenue per employee, error rate, CSAT, sales velocity, onboarding time, renewal rate.

## The "founder makes the implicit explicit" list

Isenberg's checklist of organizational implicit-knowledge that has to be surfaced and codified for agents to operate:

- Refund policy — and when it gets broken.
- Lead qualification criteria.
- Tone with angry customers.
- What should never be automated.
- Which actions require approval.
- What "good" looks like; what "dangerous" looks like.
- Source-of-truth data system; conflict-resolution between disagreeing systems.
- Feedback loop for agent corrections.

> "Everyone wants the magic. Nobody wants to clean the kitchen. But the kitchen is the company."

## Concrete vertical examples (verbatim)

- **Home services co.** Every inbound request classified. Quotes generated from structured pricing rules. Technicians get job summary before arrival. Customers get proactive updates. Reviews personalized. Missed appointments trigger recovery workflow. Operational patterns feed back into routing, pricing, staffing.
- **Insurance brokerage.** Agents gather documents, pre-check submissions, compare policies, flag missing details, draft client explanations, prepare renewals, monitor for changes. Humans build trust and handle complexity.
- **Recruiting firm.** Agents source, enrich, match against role, draft outreach, summarize interviews, check references, update pipelines, flag unusual candidates. Recruiter stops being a data janitor and becomes a relationship closer.

> "These are not sci-fi companies. These are normal businesses with the guts rebuilt."

## Memorable lines

- "An AI-native company is not a company that uses AI. It is a company that has been rebuilt so AI can actually operate inside it."
- "The company has made itself legible to machines."
- "Most companies have spent twenty years buying software, but they have not spent twenty years designing an operating system. They have a pile of tools, not a machine."
- "A lot of work is not really 'work.' It is organizational friction wearing a fake mustache."
- "Not AI as a party trick. AI as organizational metabolism."
- "When a market is loud, people assume it is mature. But noise is not maturity."
- "Most people are still asking, 'How do I use AI at work?' The better question is, 'How do I build a company AI can work inside?'"
- "They will make incumbents look like they are running Windows 95 with a nicer login screen."

## Why this is in the vault — the 2026-05-11 triangulation

Third major primary-source signal on the same day (2026-05-11) converging on the same thesis from three distinct vantage points:

| Vantage | Source | Claim |
|---|---|---|
| **Supply side** | [[08 Reference/2026-05-11 OpenAI - Launches the Deployment Company - Digest\|OpenAI Deployment Company]] | $4B+ JV with TPG/Advent/Bain/Brookfield to scale Forward Deployed Engineers + Tomoro acquisition. |
| **Demand side (CEO operator view)** | [[00 Raw/X Posts/2026-05-11 Aaron Levie - Professional Services and FDEs for Agents\|Levie — Professional Services and FDEs for Agents]] | Services/consulting wave from agents will exceed analog-to-digital and on-prem-to-cloud combined; the technical surface is data + identity + context + eval + change mgmt. |
| **Demand side (operator hiring view)** | [[00 Raw/X Posts/2026-05-11 Aaron Levie - AI Automation Engineers at Box\|Levie — AI Automation Engineers at Box]] | Box itself hiring internal FDE-equivalents partnered with functional SMEs. |
| **Builder/investor view** | This essay | "Only ~1,000 truly AI-native companies on earth"; the opportunity is to rebuild boring categories with agent-native operating models; new winners will look like services businesses with software margins hidden inside. |

All four signals describe the same underlying transition: **the unit of value capture is the workflow, the go-to-market motion is forward deployment, the binding constraint is the customer's organizational legibility, and the company structure that wins is the small-team-with-agent-fleet rebuild rather than the incumbent retrofit.** Strong reinforcement for [[09 Implications/Migrate Workflows Not Seats]] and a clean extension of the services-as-software thread in [[02 Briefings/Daily/2026-04-25 Sequoia - Services The New Software - Digest]].

## What's distinctive about Isenberg's contribution

Beyond the convergence with Levie/OpenAI, this essay adds three things to the vault:

1. **The "AI-native vs. AI-assisted" vocabulary.** Cleanest framing yet for the edges-vs-center distinction. Useful as a screening question for any company claim.
2. **The "boring vertical with software margins hidden inside" investor thesis.** Where Levie talks supply/demand for the services wave, Isenberg specifies the new-company-formation strategy: pick a narrow, boring, profitable, fragmented industry; rebuild its operating model around agents; let the financial profile look like services on the surface and software in the cash flow.
3. **Revenue-per-employee as the dominant external metric.** Not just margin or growth — a structural signal that an organization has actually rebuilt for the new model rather than bolted AI on top. Worth tracking explicitly across the vault's company notes going forward.

## Open questions

- The "1,000 AI-native companies" claim is intentionally a hunch — what would a concrete operationalization look like? Possible test: companies where >50% of customer-facing workflows have a defined agent action with human-supervised review points, and revenue/employee is >2× their category median.
- Which boring vertical gets disrupted first? Insurance and recruiting (both named by Isenberg) have high knowledge-work labor cost and structured outputs — likely early. Home services is messier (physical world) but has clean transactional data.
- Does the incumbent-can't-retrofit claim hold? Or do a small number of incumbents successfully rebuild their OS, and if so, what's the precondition (founder still in seat? small enough? cloud-native already?)?
- How does the "joint human + agent identity/entitlements" layer (raised by Levie) get built? Standalone market or feature of an existing IDP?

## Capture caveat

URL not provided. Verbatim text was pasted into the vault session and is preserved in the quotes above; the full essay body is also recoverable from chat history if a primary-source check is needed. If you have the original URL, drop it in and I'll update the `source` field.
