What is agent-to-agent billing?

Agent-to-agent billing is a model where AI agents on the buyer's side (accounts payable) and the seller's side (accounts receivable) negotiate invoice details, disputes, and payment terms directly with each other, with humans only stepping in for approvals or exceptions. It's the same pattern Aegis is building for healthcare claim adjudication, where provider-side and payer-side agents settle reimbursement amounts without manual back-and-forth.

How is AI changing revenue cycle management (RCM) in healthcare?

Insurance companies have been using AI to adjudicate claims for years, which is why initial denial rates now sit at 15% and 54.3% of denials are overturned on appeal (Premier Healthcare, 2022). AI on the provider side, like what Aegis is building, closes that asymmetry by automatically generating appeals, structuring unstructured patient data, and submitting prior authorization requests. The goal is to reduce the 80-90% of human RCM work that's currently spent on repeatable, rules-based tasks.

Why aren't voice agents the future of billing or collections?

Voice agents bolt AI onto an existing manual process (phone calls) instead of replacing the process. In healthcare, insurers hang up the moment they detect an AI on the line. In B2B billing, customers don't want AI calling them about late invoices. The structural fix is preventing the dispute upstream: cleaner contracts, better invoice-to-PO matching, and reconciled usage data. The phone call is a symptom, not a feature.

What is a "20X company"?

The "20X company" is a framing popularized by Y Combinator CEO Garry Tan: a startup where every internal function (sales, support, ops, legal, finance) is automated with AI, allowing a team of 3-12 people to match the output of a company 20 times their size. Roughly 25% of YC's current batch ships codebases that are 95% AI-written, and the cohort is the fastest-growing in YC's history.

How should B2B finance teams prepare for agent-native billing?

Three moves: (1) audit which billing decisions — dunning, disputes, revenue recognition adjustments — are still rule-based human work, since those go agentic first; (2) stop buying tools that replicate manual workflow and start buying tools that eliminate it; (3) when evaluating billing vendors, ask specifically what happens when a buyer's AP agent disputes one of your invoices — if the answer involves a human ticket on your side, the architecture is already obsolete.

The Agent-to-Agent Future of Billing Is Already Here. Healthcare Saw It First.

Shrinija Kummari

May 5, 20269 min read

Futuristic digital illustration of a tablet with a 3D professional figure and medical organ icons emerging from the screen, on a dark HUD-style tech background.

U.S. hospitals spend $19.7 billion every year fighting denied insurance claims. More than half of that — $10.6 billion — is wasted on appealing claims that should have been paid the first time. The reason the number keeps climbing is simple, and it has nothing to do with healthcare.

Insurance companies started using AI to deny claims faster than hospitals could appeal them. The provider side never caught up.

That asymmetry is the leading indicator for what's about to happen to every B2B billing relationship. I sat down with Krishang Todi, CEO of Aegis (YC 2025), on the JustPaid podcast last week. Krishang is building AI agents that handle insurance claim denials end-to-end — and the way he frames the future of revenue cycle management is the clearest preview I've heard of where SaaS billing is heading next.

The Denial Loop Is Already Agentic — On One Side

Krishang's framing of the healthcare problem is precise: insurers deployed AI for adjudication years before providers had any AI defense. The result is a one-sided arms race.

"Insurance companies have been using a lot of AI technologies to adjudicate claims, leading to more and more denials. The traditional infrastructure that hospitals have in place to fight back has started failing down."

The numbers back him up. According to Premier Healthcare's national denials survey, 15% of all claims submitted to private payers are now initially denied. 54.3% of those denials get overturned on appeal — meaning the claim was valid all along. Hospitals spend an average of $43.84 per claim just to contest a denial that the insurer already knew should have been paid.

This is the same shape forming in B2B SaaS. Procurement teams at large buyers are already using AI to scrutinize vendor invoices, catch usage discrepancies, and dispute charges automatically. Most finance teams on the seller side are still chasing payments in spreadsheets and Slack DMs. The seller-side response is manual; the buyer-side challenge is automated. That gap is where revenue leaks.

What "Agent-to-Agent" Actually Looks Like

Krishang's vision for Aegis isn't "AI helps the appeals clerk work faster." It's the appeals clerk going away.

His description of the end state:

"You have AI agents on both the provider side and the insurance side that figure out amongst themselves what is an appropriate amount that should be reimbursed. Our agents can automatically file prior authorization requests, submit claims, and if there's a denial, automatically fight it back."

He estimates 80-90% of the humans currently doing this work won't be needed.

This is the shape of the change, and it generalizes. Any billing dispute is structured data plus rules plus a back-and-forth negotiation. That's the exact workload LLMs are good at, and it's exactly how B2B billing disputes work today — from contract interpretation to usage true-ups to revenue recognition adjustments.

Workload	Today	In 24 months
Invoice dispute response	AR analyst, 2-5 days	Agent, minutes
Contract-to-invoice reconciliation	Manual diff in Excel	Agent extracts + flags
Dunning sequence escalation	Rule-based email tool	Agent negotiates payment plan
Revenue recognition adjustments	Controller + auditor review	Agent proposes, human approves

The humans don't disappear. The work they do shifts from execution to oversight.

Why Voice Agents Aren't the Endgame

The most useful thing Krishang said about where AI in healthcare is overhyped:

"As soon as the insurance company picks up the call and realizes it's an AI, they would cut the call."

This is happening across healthcare right now. Voice agents calling insurers to check claim status get hung up on the moment they're detected. The "innovation" was bolting an AI mouth onto a broken phone-tree process. The phone tree was the problem.

The same lesson applies to B2B collections. Nobody — buyer or seller — wants to be called by an AI about a late invoice. The win isn't "AI calls your customers." The win is preventing the dispute from existing in the first place: the contract terms were ambiguous, the invoice didn't match the PO, the usage data wasn't reconciled. Fix that upstream and the call never has to happen.

This is why we built JustPaid's AI contract extraction and reconciliation layer before we built Reminder AI. The reminders are downstream. The disputes start at the contract.

The "20X Company" — Why Lean Teams Will Win Billing Too

Krishang made a comment that's worth reading twice:

"We're able to build this entire software with a team of just three people. We've mastered understanding how AI works. So we are more likely to provide you a more autonomous solution than any of these other companies who have hundreds of people on their payroll."

He was referencing Garry Tan's "20X company" framing — the idea, popular inside YC right now, that a startup where every internal function is automated with AI can match the output of a company 20 times its size. According to YC, about 25% of its current batch ships codebases that are 95% AI-written. CNBC reported earlier this year that the current YC cohort is the fastest-growing in the fund's history, and AI is the reason.

The implication for B2B finance teams is uncomfortable but obvious: the AR org of the future isn't 15 people. It's 3 people running a system. Not because of layoffs — because the work itself collapses. The companies that figure this out first will operate with structurally lower G&A than their competitors, which compounds into pricing flexibility, which compounds into market share.

If you're a finance leader, the question isn't "how do I add AI to my team." It's "what does my team look like when the work is 20x smaller."

What This Means for Finance Teams Right Now

Three concrete moves for B2B finance and RevOps leaders this quarter:

1. Audit which billing decisions are still rule-based human work. Dunning logic, dispute responses, revenue recognition adjustments, usage true-ups. These are the first to go fully agentic. If a junior analyst can do the task with a runbook, an agent can do it for $0.04 per call.

2. Stop buying tools that replicate manual workflow. The market is full of dashboards that "give your team visibility" into things a human still has to act on. Buy tools that eliminate the workflow, not tools that make the workflow prettier.

3. Pick vendors building toward agent-native, not chatbot-bolted. Ask any billing vendor a specific question: when a buyer's AP agent disputes one of your invoices, what happens? If the answer involves a human on your side opening a ticket, the architecture is already obsolete.

The Takeaway

Krishang is solving denial appeals because that's where the asymmetry is most painful right now. But the structural insight — agents on both sides of a financial transaction, negotiating with each other — is the same insight reshaping every billing relationship in B2B. Healthcare is just the canary.

The founders and finance leaders who internalize the agent-to-agent shape now will spend the next five years building a different cost structure than everyone else. The ones who wait will be appealing their own version of $19.7 billion in lost revenue, one ticket at a time.

If you're rebuilding your AR motion for an agent-native world, book a JustPaid demo — that's the world we're building for.