Why AI Agents Are Saving the Fax Machine, Not Killing It

# Why AI Agents Are Saving the Fax Machine, Not Killing It

Nish Hari runs Predoc, a company building AI agents to retrieve medical records. By every instinct of the AI founder playbook, he should be rooting for fax to die. He's not.

In a piece for Second Opinion Media, Hari made the counterintuitive case that AI agents will extend the life of the fax machine rather than end it. The argument is worth taking seriously — not just for healthcare founders, but for anyone building agentic systems into legacy-heavy industries.

The fax machine is not a relic. It's infrastructure.

In U.S. healthcare, an estimated 9 billion faxes are sent annually. Hospitals, clinics, and insurance providers have built decades of workflow, compliance, and trust around fax. It is HIPAA-compliant by default. It leaves a paper trail courts accept. It doesn't require the receiving party to have the same software, API, or vendor relationship as the sender.

In other words: fax has the properties that matter most in regulated, high-stakes environments. Interoperability. Legal durability. Ubiquity.

EHR systems like Epic and Cerner have tried for years to create unified digital pipes. They haven't replaced fax. They've coexisted with it.

What agents actually do with fax

Hari's insight is structural: agents don't need clean digital inputs to function. They need *retrievable* information. Fax, phone calls, PDFs — these are all surfaces an agent can read from, write to, or orchestrate around.

Predoc's agents reportedly call provider offices, navigate phone trees, request records, and process what comes back — including faxes. The agent wraps around the interface rather than replacing it.

This is the pattern that most AI commentary misses. The default assumption is: AI replaces the old process. The more accurate picture, at least in the near term, is: AI agents become the most capable users of the old process.

A human medical records coordinator can handle maybe 20-30 record requests per day. An agent working phone and fax systems can run hundreds in parallel, without demanding that hospitals rewire their infrastructure.

Why this matters beyond healthcare

Fax is a stand-in for a broader category: sticky legacy interfaces that persist because switching costs are too high, regulation is too heavy, or the installed base is too large.

Think about: - Email inboxes at large enterprises (still the dominant async communication layer) - PDF-based procurement and invoicing workflows - Phone-based customer service trees in financial services - Physical mail in legal and government contexts

In every case, the prediction has been "this will be replaced by something better." In most cases, the reality is that the interface survives while a layer of intelligence gets built on top of it.

AI-native companies that are scaling fastest aren't always the ones with the cleanest tech stacks. Some of them are winning precisely because they're willing to operate inside the mess — using agents to make old infrastructure fast, accurate, and cheap to run.

The solo founder implication

For a one-person startup, this is a useful reframe. You don't need to wait for industries to modernize before building agents into them. The friction *is* the moat.

If a sector still runs on fax, phone calls, and PDFs, it almost certainly has: - No dominant AI-native player yet - High tolerance for paying someone to handle the overhead - Decision-makers who are exhausted by the status quo

An agent that works *with* the existing interface — rather than demanding a new one — can get deployed in weeks, not quarters. You skip the integration fight. You skip the procurement committee debating whether to migrate systems. You just show up, handle the fax queue, and charge for outcomes.

This is exactly the kind of wedge that produces high revenue per employee early. The agent does the volume work. The founder manages the edge cases, improves the system, and expands scope.

What "durable interfaces" tell us about agent design

Hari's framing — that phone and fax are "the most durable interfaces" — is a useful design constraint.

If you're building an agent for any industry with regulatory weight, ask:

1. What interface has survived every modernization attempt in the last 20 years? That's where volume lives. 2. What does compliance require the interface to do? (Audit trail, consent, delivery confirmation — fax satisfies all three in healthcare.) 3. Can an agent operate on this interface without requiring the other party to change anything? If yes, adoption resistance drops to near zero.

The best AI agents in regulated industries won't announce themselves. They'll quietly become the most reliable participant in a workflow that everyone assumed was stuck.

The counterintuitive competitive edge

There's a second-order point worth naming: the companies that lean into legacy interfaces are harder to displace than the ones trying to replace them.

If Predoc builds deep capability in navigating phone trees and processing faxes, they become more valuable as those interfaces persist — not less. If a competitor comes along with a "clean" digital-only solution, they still can't reach the 60% of provider offices that won't adopt a new portal.

The agent that speaks fax wins the accounts the clean-tech competitor can't touch.

For founders building AI-native companies in healthcare, legal, insurance, or government, this is the strategic question: are you building for the world as it should be, or the world as it is? Hari chose the latter. Predoc is growing because of it.

The takeaway

AI agents are not a force for technological purity. They're a force for operational efficiency — and operational efficiency often means working with whatever interface the industry already trusts.

Fax machines will outlive most of the startups that predicted their death. The founders who understand that early will build agent systems that age well, integrate fast, and become genuinely hard to replace.

The fax isn't the problem. It's the opportunity.

---

Is your company eligible? Submit to the leaderboard → onepersonunicorn.co/submit