Two weeks ago we covered Xero and Intuit signing multiyear deals with Anthropic and asked which side your platform was picking. This week, the answer came back: all of them. Every major ledger platform now has at least one frontier AI partnership — and two of the three have two. Meanwhile, Anthropic had its roughest week yet, Google handed the industry a free escape hatch, and one person quietly proved that the "billion-dollar company with no employees" isn't a thought experiment anymore.
Every platform picked every side
The Intuit-OpenAI partnership we first covered in November has shipped. TurboTax, Credit Karma, QuickBooks, and Mailchimp now live inside ChatGPT as embedded apps. Users can run forecasts, prep taxes, and manage payroll through conversation. Intuit also unveiled a "System of Intelligence" with dedicated AI agents for accounting, customer management, and finance functions. This runs parallel to the Anthropic partnership we covered two weeks ago — Intuit is building on both.
Xero's doing the same thing. Its JAX superagent is being rebuilt on OpenAI for bank reconciliations, data entry, and compliance automation. The Anthropic partnership handles cash flow predictions and task automation. Two vendors, two model providers, two bets hedged.
And Sage — which we don't cover as heavily — just closed the gap. They shipped a Finance Intelligence Agent for natural language queries across Intacct data, an AI-powered import agent, and a Snowflake integration with zero ETL. Sage hasn't announced a frontier AI partnership comparable to what Intuit and Xero have (both have OpenAI and Anthropic), but the product capabilities are converging fast. Sage Future Conference runs April 28–30 in San Francisco — watch for that announcement.
The pattern is what matters here. If every platform has native AI doing bank recs, data entry, and compliance checks, the commodity layer thesis holds: the competitive asymmetry between platforms narrows, and your value as a practitioner shifts entirely to what sits above the automation — interpretation, judgment, client communication, and advisory. "Platforms picking sides" was the story two weeks ago. "Platforms picking all the sides" is the update. The infrastructure is becoming interchangeable. Your expertise isn't.
Anthropic's roughest week — and the open-source counter-move
On March 31, a 59.8 MB source map file was accidentally included in a Claude Code npm release, exposing roughly 513,000 lines of unobfuscated TypeScript and 44 unreleased feature flags — essentially a product roadmap in code form. Anthropic issued about 8,100 DMCA takedowns on GitHub, then retracted most of them, creating a secondary PR problem. This came 10 days after the Mythos model leak through an unsecured CMS. Two security failures in under two weeks from the same company.
Then on April 4, Anthropic ended flat-rate subscription access for third-party agentic tools like OpenClaw, forcing users to pay-as-you-go. If your firm built workflows through agent orchestrators using a Claude subscription, those costs just became variable. That's the SaaS-to-consumption pricing shift we've been tracking since March — except now it's hitting your own AI subscription, not someone else's.
Here's the counter-move. On April 2, Google DeepMind released Gemma 4 — four open-weight models under the Apache 2.0 license. No MAU caps. No acceptable-use restrictions. Full commercial freedom. The 31B model ranks third on the Arena AI text leaderboard, supports 256K context, native vision and audio, and 140-plus languages. If frontier-class models are free to embed commercially, any software vendor — including accounting platforms and niche CAS tools — can bring high-quality AI processing in-house at near-zero model cost.
These three stories tell one narrative. Building on a single AI vendor carries real risk — security risk, pricing risk, dependency risk. But the escape hatch just got wider. The differentiation shifts from which model you're paying for to how you design workflows and integrate data around it. Vendor dependency, made concrete.
The one-person unicorn is here
Matthew Gallagher launched Medvi — a GLP-1 telehealth startup — from his Los Angeles home with $20,000, no employees, and 12-plus AI tools. First-year sales: $401 million. 250,000 customers. A 16.2% net profit margin. Now tracking to $1.8 billion in 2026 revenue with a headcount of two — Gallagher and his brother.
This validates predictions from Dario Amodei and Sam Altman about the one-person billion-dollar company. Solo-founded startups now represent 36.3% of all new ventures. The abstract "AI changes everything" claim just got a receipt.
For CAS, this inverts a fundamental assumption. The advisory opportunity isn't helping this client scale a team. It's helping one person manage the financial complexity of hyper-growth on minimal staff — tax implications across 250,000 customer transactions, revenue recognition, entity structure, multi-state compliance. The threat is CAS practices that still measure their own value by the hour when a solo founder generates more revenue than most of their client base combined. Saturday, we pull apart what a $1.8 billion solo founder actually needs from an accountant — and why most small CAS practices will lose these relationships before they realize they had them.
Quick hits
OpenAI closed a $122 billion round at an $852 billion valuation. Amazon, Nvidia, and SoftBank each put in between $30 billion and $50 billion. Revenue is $2 billion per month with 900-million-plus weekly active users. The "wait and see" strategy grows riskier by the quarter.
Codex goes pay-as-you-go. OpenAI added usage-based Codex seats for Business and Enterprise — no rate limits, billed on tokens. Two million weekly users, 6x growth since January. Another data point in the consumption pricing shift.
GPT-5.4 mini launched. Near-flagship performance at roughly 6x lower cost. The mini-to-flagship gap continues to narrow — routine AI tasks should be running on minis, not flagships.
Cursor 3 shipped. Complete rebuild from code editor to agent orchestration platform. Agents Window for running many agents in parallel, Design Mode, cloud-local handoff. No longer a VS Code fork. Leading indicator of where all professional tooling is headed.
Where this leaves you
Three patterns running in the same direction this week. Platforms converging means the tools get smarter regardless of which one you're on — and the value of the tool itself drops relative to the value of what you do with it. Vendor risk getting real means you need to know what you're building on and what your fallback is. And a solo founder hitting $1.8 billion means the old metrics — headcount, hours billed, bodies in seats — are losing their grip on what "scale" means.
The common thread: the infrastructure layer is commoditizing. Models, platforms, even entire teams — all compressing toward cheaper, faster, fewer. What doesn't commoditize is your judgment, your client relationships, and your ability to interpret what the machines produce. That's where the margin lives now. Thursday, we dig into what that commoditization does to your firm's valuation — the data says most practice owners approaching retirement don't know which side of the K they're on.
If you're ready to move from watching the infrastructure layer compress to actually building workflows in the gaps between your platforms, start with the free AI in the Gaps Toolkit — 100 specific workflows across 20 categories, scored by effort and impact, with implementation context for every one. You don't need API access, vendor integrations, or months of planning. Pick three workflows. Build them. Then pick three more. Download the toolkit here.