Neofirms — AI-Native Professional Services as a New Category

The definition

A Neofirm is an AI-native professional services firm with three characteristics:

1. AI does the bulk of the work. The firm’s operational backbone is AI agents running foundation models, often with vertical-specific fine-tuning, proprietary prompts, and domain-specific tool integrations. The AI is the primary worker. Humans handle exceptions.

2. Humans own quality, judgment, and accountability. A small team of domain experts — often orders of magnitude smaller than an equivalent traditional firm — reviews, validates, corrects, and signs off on the AI’s output. The humans are also the vendor-of-record that takes legal responsibility for the deliverable.

3. The firm sells outcomes, not tools. Customers pay per completed deliverable (contract reviewed, return filed, policy written, codebase modernized), not per seat or per API call. The firm captures economic value proportional to the work it completes, not proportional to how many people access its software.

The term was popularized in 2026 by Ryan Daniels, co-founder of Crosby AI, a legal services Neofirm that combines software, AI, and attorneys to deliver legal work to customers at a fraction of traditional BigLaw cost.

The examples

Crosby AI (legal). Combines AI agents for contract review, drafting, and research with attorneys for final review and strategic judgment. Customers get the output of a law firm without directly hiring one.

Harper (insurance). AI-native insurance brokerage. Rather than selling insurance software to brokers, Harper IS the broker. Customers get matched to policies through Harper’s AI + human team, and the firm captures the broker’s margin rather than licensing fees.

WithCoverage (insurance). Similar model in a different segment of the insurance market. AI does the routing, matching, and initial recommendation; humans confirm and close.

Mechanical Orchard (software modernization). Sells completed COBOL-to-modern-stack migrations rather than migration tools. AI + engineers produce working modernized systems as the deliverable.

The pattern repeats: in each category, traditional firms were expensive and slow. Pure software players tried to sell tools to those firms but captured only a fraction of the underlying budget. Neofirms cut out the middle. They skip the firm and become one themselves, with AI economics.

Why Neofirms emerge now

Three conditions have to all be true for a Neofirm to work in a category:

1. AI is capable enough to do the bulk of the work. The output of modern foundation models has to be 80%+ of the quality of a human professional at the same task. This threshold crossed in late 2024 for many knowledge-work categories — legal drafting, tax prep, insurance routing, medical triage, basic software engineering. Below that threshold, AI was a productivity tool. Above it, AI is the worker.

2. The work cannot yet be delivered without human oversight. If AI were reliable enough to deliver without review, the category wouldn’t need a Neofirm — it would be a pure software play. The Neofirm model depends specifically on AI being 80-95% reliable, which is both too good to ignore and too risky to ship without humans in the loop. This band is where Neofirms win.

3. Traditional firms are expensive and slow. If an incumbent professional services firm is already delivering the work efficiently at competitive prices, there’s no opening. Neofirms thrive specifically in categories where traditional firms charge $300-$1,500/hour for work that mostly follows patterns — exactly the categories where AI can compress the cost structure by 50-80%.

As AI reliability improves (conditions 1 and 2 drift toward “AI can do it all”), Neofirms will probably evolve into either pure software companies (dropping the human layer) or AI-driven platforms that white-label services to traditional firms. The current Neofirm category is transitional, not permanent. But for the 3-7 year window where AI is “most of the work” but not “all of the work,” the Neofirm model captures margin that neither pure software nor traditional firms can.

The economic argument

A traditional professional services firm has roughly 20-30% net margin with 70-80% of revenue flowing to partner and employee compensation. A pure software company selling tools to that firm captures 5-15% of the firm’s total budget as software license revenue.

A Neofirm in the same category can capture:

• The client budget that used to flow to the firm (more of it, because the client pays per outcome not per hour)
• The labor cost savings from AI doing 80% of the work
• The software margin from owning the vertical stack

Ballpark: if a traditional firm charges $10,000 for a contract review and the Neofirm can deliver the equivalent result for $3,000 using AI + a junior attorney reviewer, the Neofirm captures $3,000 of revenue at perhaps $500 of actual cost (mostly the human reviewer’s time). That’s a 75-85% gross margin on a transaction that used to be 25-30% margin for the incumbent. The Neofirm wins on cost, the customer wins on price, and the margin gap accrues to the Neofirm.

The catch: this only works at scale. A Neofirm needs to run enough transactions for the AI/software leverage to dominate the cost structure. Single-engagement services (bespoke M&A advisory, one-off strategic consulting) don’t fit the model because each engagement requires enough human judgment that AI leverage never takes over.

Where Neofirms fit best

Categories where Neofirms are most likely to succeed share these traits:

• Work follows patterns. Contract review, insurance routing, tax prep, standard legal filings, routine medical triage, basic financial analysis. Highly bespoke work doesn’t fit.
• Output is measurable and verifiable. The deliverable can be checked against clear criteria. “Is this contract compliant?” is measurable; “Is this strategy the right one?” is not.
• Current incumbents charge hourly rates that far exceed the actual operational cost. The margin gap that Neofirms can capture is proportional to how inflated the incumbent’s pricing is relative to the underlying labor and technology cost.
• Liability and trust can be credentialed. Customers will hand over sensitive work only if the Neofirm has the legal standing (licensed attorneys, registered brokers, certified practitioners) to take responsibility. The human layer provides the liability coverage the customer needs.
• Results can be delivered at distance. Professional services that require physical presence — surgery, in-person negotiation, live hearings — don’t fit. Most knowledge work does.

Categories where Neofirms struggle: any category where the work is deeply bespoke, where liability cannot be credentialed remotely, where the incumbent’s pricing is already efficient, or where the output of AI is not yet good enough for 80% of the work.

The counter-argument

Two honest objections to the Neofirm thesis:

1. “This is just a staffing agency with AI tools.” The counter-argument says Neofirms are just traditional professional services firms with better tools. Over time, traditional firms will adopt the same AI tools and the Neofirm advantage will disappear. The response: this is probably true for traditional firms that move fast. The large majority will lose market share to Neofirms before they catch up. The transition window is the opportunity.

2. “Liability won’t transfer cleanly.” The counter-argument: regulators, insurance underwriters, and customers will not accept “the AI made a mistake, but our human reviewer signed off” as a defense. Professional liability requires clear accountability, and if the AI is doing 80% of the work, the human reviewer’s liability coverage may not scale. Response: this is a real risk and probably limits how far AI can go in regulated professions before the liability framework has to adapt. Neofirms should be thought of as early entrants in a category that will eventually need new legal/insurance/regulatory infrastructure to fully mature.

2026-09 Update — The Primary Source, and the Labour TAM

This note has been citing Ryan Daniels second-hand through Tanay Jaipuria. The primary source is now in the vault: The White Collar Revolution Is Here — Daniels’ own argument for the category he named, which supplies a structural mechanism the note’s “why now” section does not have.

The R&D argument. America’s top 100 law firms made a combined $69 billion in profit in 2025, more than Google’s 2025 R&D budget, and every cent was paid out to partners as compensation. That is not greed; it is the partnership form working as designed. Partnerships pay out profits annually as a dividend and legally cannot sell equity to outside investors, so they have no vehicle for funding multi-year technology work and no incentive to try.

This is a better answer to “why can’t incumbents just adopt the same AI tools” than the counter-argument section’s current response, which relies on incumbents being slow. The claim here is that they are structurally unable to capitalize R&D, not merely late. Figures are Daniels’ own and he is selling the category, so treat the $69B as reported rather than verified.

Two operating consequences he names:

• A 50/50 split of practitioners and engineers or AI researchers. Not a services firm with an engineering team attached, and not a software company with domain advisors. Equal weight, which is what makes the firm able to decide what to build.
• Billing per document rather than per hour. This is the piece that makes the R&D reinvestment coherent. Under hourly billing, a lawyer spending an afternoon writing agent benchmarks is lost revenue. Under per-document billing, that afternoon buys margin next year, which he calls clarifying.

The credentialing prediction: “Instead of choosing your banker based on their alma mater or firm prestige, you will ask to see their benchmarks.” If evals become the measure of professional output, the choice between a $3,000/hour partner and a $1,000/hour one becomes an informed one. He concedes the limit himself — some of what you pay for is reputation, and not everything top professionals do can be quantified.

The labour TAM this note has been assuming

Services - The New Software puts numbers on the opportunity the economic argument above treats qualitatively. The governing ratio: for every dollar spent on software, six are spent on services.

The wedge rule is sharper than this note’s “work follows patterns” criterion: start where the work is already outsourced. An existing outsourcing contract proves the company accepts external delivery, provides a budget line that substitutes cleanly, and means the buyer is already purchasing an outcome. Replacing an outsourcing contract is a vendor swap; replacing headcount is a reorg. Crosby started with NDAs for exactly this reason.

All figures are one investor’s estimates in a public thesis post. Ranges, not measurements.

The workflow is the compounding asset, and it can be given away

An AI-native marketing agency open-sourced its entire workflow library (ericosiu/ai-marketing-skills, 2,400+ stars self-reported) — skills for finding ICPs, mining sales calls, scoring content ideas, building outbound lists, auditing landing pages, finding SEO gaps, turning podcasts into content, modelling finance tradeoffs.

The operator’s framing: “That’s how service work starts compounding like software.” Publishing the library rather than guarding it is the thesis stated in the open. If the compounding asset were the workflow text, giving it away would be suicide. The bet is that the asset is the practice of running, reviewing and improving the workflows against real clients, which a repo does not transfer. That is testable, and it is the same bet the Neofirm model makes about proprietary process.

Sources added 2026-09-07

• ericosiu/ai-marketing-skills

Related Notes

• Full Stack Down vs Full Stack Up - Two Directions for AI Application Vertical Integration
• Run the Business vs Do the Work - The AI-Era Vertical SaaS Shift
• System of Action - Evolution Beyond System of Record
• Healthcare Admin Automation - Data Transformation and Vertical Workflows
• Horizontal Platform to Vertical Specialization - Enterprise Credibility Pivot
• Enterprise AI Moat - Liability and Security Gravity
• AI + Business Model
• Absorb Automate Unbundle - Three Phases of Technology Deployment — neofirms are the unbundling phase outcome for professional services
• Tanay Jaipuria - AI Applications and Vertical Integration — Tanay Jaipuria
• The 003 Problem - Labor Budgets Evaporate Instead of Redirecting to AI — Neofirms partially escape the 97% repricing trap because human-in-the-loop accountability has a price floor
• Services First, Productize Second — services as a discovery stage rather than a destination, and the ceiling that argues for productizing
• The Solo Operator Playbook - Five Primitives — the one-person version of the same structure
• Harvey CEO Winston Weinberg on Prioritization and Saying No — how one of these firms is actually run day to day