I thought this was going to be a slow news week. I was wrong.
I sat down on Monday expecting a quiet week. Maybe a model update. A funding round or two. Something I could cover in a couple of paragraphs and call it done.
That lasted about an hour. And then Friday afternoon, Sequoia dropped something that changed the whole picture. Here's what actually happened — starting with the one that matters most.
Sequoia Capital just named your practice a disruption target
Late Friday, Sequoia Capital — the firm behind Apple, Google, and Stripe — published a thesis called "Services: The New Software." The argument: the next trillion-dollar company won't sell you a tool. It'll do the work. And they name accounting explicitly.
Their numbers: the US accounting and audit market represents $50–80 billion in outsourced spend. The profession has lost 340,000 accountants in five years. Seventy-five percent of CPAs are nearing retirement. Demand is growing while supply collapses.
And then the line that should keep you up tonight: a company pays $10,000 a year for QuickBooks and $120,000 for an accountant to close the books. The next legendary company will just close the books.
They're talking about you.
This is too big for a roundup. Tomorrow, I'm doing a full breakdown — what Sequoia's framework actually says, where CAS practices are exposed, and what's defensible. And on Friday, March 20 at 3pm Eastern, I'm spending a full hour going deep on this — your pricing, your staffing model, your five-year plan. Free. Register now at theaiaccountant.ai/webinar.
Now for the three stories that shaped the rest of the week. Each one is significant in its own right, and we're going to spend more time on each of them throughout the week — so make sure you're subscribed.
AI accounting just got its first unicorn
Basis — an AI-native accounting platform — raised $100 million at a $1.15 billion valuation. Led by Accel, Google Ventures, and Lloyd Blankfein personally. That makes it the first AI accounting unicorn.
Here's what they actually do. Basis runs a multi-agent architecture where a supervising AI agent coordinates specialized sub-agents for journal entries, reconciliation, and technical accounting. These aren't quick-prompt assistants. They're long-horizon agents that work autonomously on complex workflows for hours — then deliver finished work for human review.
Every output surfaces the data sources, mapping logic, and a confidence score. The accountant reviews the reasoning, not just the result.
The numbers: approximately 30% of the Top 25 US accounting firms are using Basis across CAS, tax, audit, and advisory. Customers include Boulay, Clark Nuber, Pinion, UHY, and Wiss. Reported time savings average 30%. And they recently demonstrated the first AI agent to autonomously complete an end-to-end 1065 partnership return.
Two important caveats. First, Basis is waitlist-only with no public pricing — this isn't something a 10-person practice can sign up for today. Second, the platform is built entirely on OpenAI's models. That's a vendor dependency worth noting — and it connects directly to story two.
But the signal is clear. Institutional capital isn't trickling into AI accounting. It's flooding in. Combined with Accrual's $75 million raise earlier this year, that's over $200 million in venture funding for AI-native accounting in 2026 alone.
The commodity layer isn't theoretical anymore. It's funded.
The government risk nobody's pricing in
Two stories collided this week that CAS practice owners should be watching closely. And we're going to cover this in more depth on Thursday.
First, the Pentagon officially designated Anthropic — the company behind Claude — a supply chain risk after Anthropic refused to drop its restrictions on mass surveillance and autonomous weapons applications. Within hours, OpenAI announced its own Pentagon deal. ChatGPT uninstalls surged 295%. Claude topped the US App Store.
Second, 78 AI chatbot bills are now active across 27 states. Oregon's passed the Senate 26-1. Virginia's AI Chatbots and Minors Act is on floor votes. The federal government is simultaneously pushing back on state-level regulation — sending Utah's Senate majority leader a letter saying their AI Transparency Act "goes against the Administration's AI Agenda."
And for our Canadian readers: Canada's proposed AI and Data Act died when Parliament prorogued last year. There's still no binding AI legislation — just a voluntary code of conduct and a newly appointed AI minister. The regulatory picture north of the border is wide open.
I'm not going to get political about any of this. But the risk picture is something we need to talk about. The AI tools you're building your practice on are subject to government policy decisions that are volatile, fragmented across 50 states, and changing week to week. One designation, one executive order, one state bill — and the tool you've integrated into your monthly close process could face new restrictions overnight.
The practical takeaway: build your AI workflows to be tool-agnostic. Don't hardwire your operations to a single provider. Context engineering — building your firm's knowledge in portable, structured formats — is what insulates you from vendor-level political risk. We'll unpack this fully on Thursday.
GPT-5.4 and the price compression accelerating beneath you
OpenAI shipped GPT-5.4 this week. Three things matter. A one-million-token context window — matching Claude's. Native computer use — a first for OpenAI's general-purpose models, meaning the AI can navigate software through screenshots, mouse, and keyboard. And pricing at $2.50 per million input tokens.
DeepSeek V4 is also imminent — a trillion-parameter model optimized for Chinese chips, with leaked benchmarks suggesting frontier-class performance.
The pattern is relentless. Every month, the context windows get larger, the reasoning gets deeper, and the cost drops. The $200 per employee per month AI cost benchmark we've been citing? It's starting to look high. At current pricing trends, firms could be running frontier-quality AI across their entire team for significantly less — which only widens the gap between firms that have adopted and firms that haven't.
Quick hits
Anthropic published labor market research using actual Claude usage data — not theoretical projections — measured against 800 occupations. Programming tasks at 75% coverage. Data entry at 67%. The gap between theoretical and actual AI automation is closing fast. We're doing a deep dive into what this means for CAS roles on Friday.
The SaaSpocalypse continues. Approximately $2 trillion in software market cap has evaporated since mid-January. Per-seat pricing models are under structural pressure from AI agents that don't need licenses.
Jason Staats highlighted Claude Cowork as the most consequential AI tool for accounting firms in his latest newsletter. And if you're interested in getting hands-on with Cowork for your practice, my book on exactly that topic is coming soon. Head to theaiaccountant.ai/cowork to be the first to know when it drops.
The Journal of Accountancy asked how accountants will learn new skills when AI automates the foundational work that used to train them. If you caught our coverage of the talent trap over the last two weeks, you're ahead of this conversation.
If you want this delivered to your inbox every week, sign up for our newsletter at theaiaccountant.ai. One email. The stories that matter. The CAS angle on every one of them.