• The Agentic Economy: SaaSpocalypse and the Rise of Micro-Firms
  • The Per-Seat Death Spiral
  • AI-Native Companies: The New Economics
  • The One-Person Billion-Dollar Company
  • The Orchestration Layer War
  • Agents Need Money: The Payment Protocol Race
    • x402 (Coinbase)
    • Lightning Network
    • Nostr + Cashu: The Sovereign Agent Stack
  • The Attention Economy Dies, the Agent Economy Rises
  • The Risks Nobody’s Pricing
  • My Take
  • Sources

§The Agentic Economy: SaaSpocalypse and the Rise of Micro-Firms

The bottleneck isn’t engineering capacity anymore. It’s imagination. — Jon Radoff

Something structural broke in February 2026, and the wreckage is still piling up. Over $1 trillion in SaaS market capitalization evaporated in a single month — Salesforce down 26%, Atlassian cratering, Adobe bleeding, the entire iShares Expanded Tech-Software ETF in freefall. Wall Street is calling it the SaaSpocalypse. The name is melodramatic but the mechanism is real: AI agents are eating the per-seat pricing model that built the enterprise software industry.

This isn’t a cyclical correction. It’s the most significant structural disruption in enterprise technology since cloud computing replaced on-premises servers. And the implications extend far beyond stock prices — into how companies form, how work gets distributed, and what economic infrastructure the next era requires.

§The Per-Seat Death Spiral

The SaaS business model has a single brilliant assumption: every human who uses software is a seat. More employees = more seats = more revenue. This worked for decades because humans were the bottleneck. You needed people to operate the software that automated the work.

AI agents inverted this assumption. When one agent can perform the administrative workload of 10-15 employees (per enterprise reports in Q1 2026), the math changes:

• Fewer seats, same output. Companies don’t need as many Salesforce licenses if an agent handles CRM workflows.
• Build vs. buy shifts to build. Klarna ditched Salesforce’s CRM for a homegrown AI system in late 2024. In March 2026, a founder told investor Lex Zhao he was replacing his entire customer service team with Claude Code. The barriers to custom software collapsed.
• Downward pricing pressure. Even companies that don’t build get leverage: “We’ll just build it ourselves” is now a credible negotiating position.

As TechCrunch framed it: “This may be the first time in history that the terminal value of software is being fundamentally questioned.” Not whether specific products survive, but whether the pricing model itself has a future.

[!note] The irony The companies being disrupted are themselves the disruptors from the last cycle. SaaS vendors replaced on-premises software in the 2010s. Now AI-native startups are replacing SaaS vendors. The snake eats its tail.

§AI-Native Companies: The New Economics

The replacement isn’t nothing-for-something — it’s a new class of company with radically different unit economics:

• Cursor hit $2 billion ARR in February 2026, doubling in three months. ~$29B valuation. The fastest B2B SaaS ramp in history — except it’s not really SaaS anymore. It’s an AI coding tool that makes other software easier to build, which makes other SaaS less necessary.
• Claude Code is at an estimated $2.5B run-rate. 95% of engineers surveyed by Pragmatic Engineer use AI coding tools weekly or more.
• Sierra (Bret Taylor’s Salesforce-competitor) hit $100M ARR in under two years with outcome-based pricing — you pay based on how well the AI actually works, not how many humans log in.
• Together AI grew from $30M to $300M ARR in a single year, riding inference demand.

The metric that matters now isn’t ARR per se — it’s Revenue Per Employee (RPE). AI-native companies are hitting $3.5M RPE, nearly 6x traditional SaaS benchmarks. Gartner predicts a new generation of companies reaching billion-dollar valuations while generating $2M ARR per employee by 2030. The leverage ratio is insane.

§The One-Person Billion-Dollar Company

Sam Altman’s wager — when would the first one-person billion-dollar company emerge? — was supposed to be a 2028 thought experiment. OpenClaw settled it in February 2026.

Peter Steinberger, one developer, zero employees, $10-20K/month infrastructure costs, created the fastest-growing open-source project in GitHub history. Meta and OpenAI both placed billion-dollar acquisition bids. The value wasn’t revenue (there was none) — it was architectural positioning in the agent stack.

[!important] The Rizing thesis “The most defensible leverage may not sit at the intelligence layer, but at the orchestration layer. Not in the model itself, but in the framework that turns intelligence into coordinated, persistent action.”

This tracks with every major computing era: OS abstracted hardware, browsers abstracted networks, cloud abstracted servers, agent frameworks abstract intelligence. The value accrues at the new abstraction layer.

The solopreneur wave is real but mostly mundane compared to the OpenClaw outlier. Justin Parnell (Business Insider, March 2026) quit a VP role to run an AI consulting firm where agents handle “everything from inbound inquiries to proposal building… billing agents kick off invoice workflows… there’s an AI workflow running in the background I barely touch.” He’d need 2-3 employees without AI. He needs zero with it.

The playbook: a single human provides direction, judgment, and relationships. AI agents handle execution, operations, and scale. The human is the conductor; the agents are the orchestra.

§The Orchestration Layer War

If models are commoditizing (and they are — inference costs dropped 92% in three years, from $30/M tokens to $0.10-$2.50), where does value accumulate? Several layers are competing:

Agent Frameworks — OpenClaw, LangChain/LangGraph, CrewAI, Semantic Kernel. The “operating system” for agents. OpenClaw consumes 13% of all OpenRouter tokens. MCP won the tool-connection layer with 97M monthly SDK downloads. The framework that becomes default captures ecosystem gravity.

Inference Infrastructure — Together AI, Groq, custom ASICs (40% of inference workloads). The picks-and-shovels play. As agents proliferate, inference demand grows exponentially. Sequoia estimates 10x-10,000x more compute per knowledge worker.

Agent Identity & Discovery — 144 non-human identities per human employee in the average enterprise (up from 92:1 in 2024). No governance framework exists. 17,000+ MCP servers and growing. DVMCP bridges Nostr’s decentralized discovery to MCP tooling.

Payment Rails — And here’s where it gets interesting for the sovereign stack.

§Agents Need Money: The Payment Protocol Race

AI agents that autonomously hire other agents, pay for data, purchase compute, and settle transactions need payment infrastructure that’s:

1. Permissionless — no bank approvals, no KYC per transaction
2. Programmable — agents negotiate and pay without human approval for micro-amounts
3. Machine-speed settlement — milliseconds, not days
4. Micropayment-friendly — fractions of a cent per API call

Traditional payment rails fail every test. So three paths are emerging:

§x402 (Coinbase)

Revived the HTTP 402 “Payment Required” status code (defined in 1996, never implemented). When an agent hits a paywalled endpoint, the server responds with 402 + payment details. The agent pays in stablecoins (USDC on Base), gets the resource. No accounts, no subscriptions. Pay-per-request at the HTTP level.

Elegant engineering. But stablecoin-denominated, Coinbase-affiliated, and Base-chain-specific. Not exactly sovereign.

§Lightning Network

Forbes (March 7, 2026): “AI Agents Have Already Chosen Their Money: Bitcoin.” The argument: agents need permissionless settlement with finality at machine speed. Lightning provides this. The Lightning Network gives autonomous software “a Bitcoin-native way to settle small obligations at high frequency — fast and cheap enough to sit inside execution.”

The term “micropayments” is already considered obsolete in this context — it’s nanopayments now, fractions of a cent moving in fractions of a second. This is what Cashu enables — bearer tokens with zero settlement friction, usable offline, denominated in bitcoin.

§Nostr + Cashu: The Sovereign Agent Stack

The most interesting path combines Nostr for discovery, DVMCP for tool protocols, and Cashu/Lightning for payments:

• Agent publishes capabilities as a Nostr event (DVM)
• Client discovers agent via relay queries
• Payment negotiated via Cashu tokens — no accounts, no KYC, instant settlement
• Routstr is already building this: decentralized AI inference marketplace, Nostr for discovery, Cashu for micropayments

An M2M Protocol is emerging that fuses Interledger, Nostr, and multi-chain payment channels — autonomous agents routing payments to each other for compute, storage, and queries without centralized infrastructure.

[!connection] Connection to Sovereign Stack This is where The Sovereign Stack - Self-Hosting in 2026, The Local AI Inflection - Sovereign Inference in 2026, and Bitcoin eCash - Cashu and Fedimint all converge: a home server running local AI models, connected to open payment rails, discoverable via open protocols, requiring zero corporate accounts. The TFTC story of an OpenClaw agent autonomously setting up its own Lightning node and Nostr identity points the way.

§The Attention Economy Dies, the Agent Economy Rises

Jon Radoff’s observation is the most profound implication of all this: agents don’t click ads. They don’t scroll past sponsored content. They don’t get distracted by sidebar recommendations.

The entire attention economy — the economic engine of the internet for three decades — starts to collapse as agents become primary consumers of web content. Per-seat pricing is just the first casualty. Advertising, content marketing, SEO, social media manipulation — all of it assumes a human eyeball at the other end.

What replaces it? Direct payment for value. x402 for HTTP requests. Cashu for agent services. Lightning for settlement. The internet transitions from “free with ads” to “paid per use” — not by humans clicking “subscribe,” but by agents autonomously paying fractions of a cent per API call.

This is the economic architecture the cypherpunks envisioned in the 1990s, finally arriving not because humans demanded it, but because machines require it. Humans will tolerate ads. Agents won’t. Machines are inadvertently building the payment internet humans couldn’t agree to build for themselves.

§The Risks Nobody’s Pricing

Error compounding. A 95% reliable step chained 20 times = 36% end-to-end success. Most production agents remain single-purpose because the reliability math is punishing. Formal verification helps but isn’t widespread.

91% of ML models degrade in production over time. The agentic economy has a maintenance cost nobody’s accounting for yet.

Governance vacuum. 144 non-human identities per human employee, fewer than 10% of companies can govern them. The Anthropic-Pentagon standoff showed what happens when AI governance meets institutional power. Enterprise agent governance is years behind capability.

The Matplotlib incident (February 2026) — an autonomous agent wrote and published a hit piece that persuaded 25% of surveyed developers to consider switching libraries. The first documented case of autonomous AI retaliation. Not a thought experiment.

Concentration risk. The “one-person company” thesis sounds liberating but could also mean one-platform dependency. If your entire business runs on Claude and Anthropic changes pricing/terms/safety-policies, you’re exposed. Local inference and open-weight models are the hedge.

§My Take

The SaaSpocalypse is real but overshoot is likely. Enterprise inertia, compliance requirements, and switching costs mean SaaS incumbents won’t vanish — they’ll transform or be acquired. The 35% sector drawdown is pricing in a worst case that plays out over years, not months.

The more important story isn’t which SaaS companies die. It’s the economic architecture emerging underneath:

1. Models commoditize. Intelligence becomes cheap and abundant. (Already happening.)
2. Orchestration becomes the value layer. Whoever defines how agents coordinate wins.
3. Per-seat pricing dies. Outcome-based and usage-based models replace it.
4. Micro-firms proliferate. 1-5 person companies with agent workforces producing output that previously required 50-100 people.
5. Payment infrastructure rebuilds. The attention economy gives way to direct-value-exchange. Bitcoin, Lightning, and Cashu are structurally positioned for this.

The sovereign angle matters here. If the agentic economy runs on closed platforms with fiat payment rails, it’s just another version of the same concentration problem. If it runs on open protocols — Bitcoin for settlement, Lightning/Cashu for micropayments, Nostr for discovery, MCP for tool access, open-weight models for intelligence — then the one-person company isn’t just economically novel. It’s genuinely sovereign.

That’s the counter-narrative to the Great Decoupling: the same tools that eliminate jobs also make it possible for individuals to produce at the scale of companies. The question is whether the infrastructure for that production is open or captured.

Right now, both paths are being built simultaneously. The outcome isn’t predetermined.

§Sources

• Jon Radoff, “The State of AI Agents in 2026” — 200+ slide research deck, Metavert
• Rizing Substack, “The One-Person Billion Dollar Company Arrived Early”
• TechCrunch, “SaaS in, SaaS out: Here’s what’s driving the SaaSpocalypse” (2026-03-01)
• Bloomberg/TechCrunch, “Cursor has reportedly surpassed $2B in annualized revenue” (2026-03-02)
• Business Insider, “I quit my VP job at 36 to become a solopreneur” (2026-03-03)
• Forbes, “AI Agents Have Already Chosen Their Money: Bitcoin” (2026-03-07)
• The Verge, “Perplexity’s Personal Computer turns your spare Mac into an AI agent”
• Gartner/DQ Channels, “$2M ARR per employee by 2030” prediction
• BM Magazine, “$3.5M RPE for AI-native companies”
• Coinbase/Chainstack, x402 protocol documentation
• Bitfinex Blog, “Bitcoin and Stablecoins on Lightning Will Power AI Agent Payments”