The world's attention was elsewhere this week, which is exactly when the interesting signals land. Digits priced bookkeeping on outcomes. Perplexity put a personal CFO on every client's phone. Anthropic quietly became the most financially important AI company in the world. The walls between vendor, tool, and practice started to move visibly.
Your practice is getting squeezed from both sides
On Monday, Digits announced outcome-based pricing for accounting firms. If Digits doesn't automate at least 95% of a client's transactions with zero human touch, you don't pay. The reference client — a top-400 firm at $10M in revenue — went from 75% automation at the start of 2025 to 98% by year-end. Two percent human touch. That's a staffing story dressed up as pricing.
The same day, Perplexity launched a Plaid integration across 12,000+ financial institutions — checking, savings, credit cards, loans — and branded it "your personal CFO." Free, read-only, available to every signed-in US and Canadian user. Your client can now watch money move across every account they own and ask questions about it in plain language, without talking to an accountant.
These aren't two stories. They're two sides of the same compression. Digits squeezes the firm from the cost side. Perplexity squeezes the client from the visibility side. If you're still selling monthly bookkeeping packages priced around bookkeeper hours, both numbers just got harder to defend.
One more thing on Digits. They're not just the threat. Back in October they launched Digits Connect — a free, GraphQL API exposing their Autonomous General Ledger to anyone who wants to build on it. No partner fees, no marketplace lock-in. The opposite of what Xero and QBO are doing. So Digits is either the platform that eats the mid-tier CAS practice or the one that lets you rebuild on top of it. Probably both. And if you're in Canada, you don't get to find out yet — they're US-only, though they say they're working on it. I'd love to tell you I've tried it. I can't.
In our Thursday follow-up, we dug into the economics behind this — what your practice actually sells when the platform handles 95% of the transactional work, and what your fee model looks like on the other side.
EY just showed what's possible — and Modus is showing what's coming
On Tuesday, EY rolled out enterprise-scale agentic AI inside Canvas, the platform running 160,000 audit engagements across 150+ countries. 130,000 assurance professionals now have AI agents in their daily workflow, processing against 1.4 trillion lines of journal entry data a year. Target: full end-to-end audit coverage by 2028. Accountancy Age called it "a watershed moment for audit quality" — unusually strong language from trade press.
We don't cover audit here. But the scale is the point. Institutional resistance — the line every CAS owner falls back on when the AI conversation gets uncomfortable — just cracked in the most regulated corner of the profession. If EY can do it, that line gets much harder to hold.
You don't have EY's budget or 130,000 staff to deploy agents to. So who builds the CAS version for you?
Modus just raised $85M to answer that question, and you may not like the answer. Lightspeed led, Garry Tan co-invested. Modus isn't selling audit software — it's investing in and partnering with audit firms, deploying its platform inside them, and using AI to compress the delivery cost. First acquisition: a top-200 firm with $30M+ in revenue, targeted to more than double its 2026 organic growth rate.
This pattern is coming to CAS next. Expect a similar vehicle — capital plus AI platform plus rollup — targeting CAS practices in the $5M–$25M range within 12 months. The acquirer landscape now includes software-first buyers with very different margin expectations. EY showed the blueprint. Modus is coming for you.
Your smallest clients may not need you at all
There's a tail to the Perplexity story. If a sole trader can get cashflow visibility and spending analysis from a free AI agent watching their bank feed, what exactly are they paying an accountant for?
This isn't hypothetical. In New Zealand, Hnry has been eating the sole-trader market with a pay-as-you-earn model — 1% of gross, capped at NZ$1,500 a year. Hnry doesn't just handle visibility. It calculates, withholds, and files tax out of the bank feed. Perplexity isn't there yet, but filing is the obvious next step.
The honest read: the $50-per-month bookkeeping client was never profitable anyway. Those clients will migrate to AI-native tools over the next 24 months whether CAS firms are ready or not. This isn't the trend accelerating — it's the trend becoming visible.
Anthropic is quietly winning the vendor durability argument
While attention was elsewhere, Anthropic's run-rate revenue crossed $30 billion — up from $9 billion at year-end 2025 and $14 billion at the February Series G. 3x growth in four months. 80% enterprise. Over 1,000 business customers each spending $1M+ per year on Claude. For the first time, Anthropic passed OpenAI (~$24B).
Two product moves landed the same week. Microsoft 365 added Claude as a first-class citizen across every paid plan — read-only access to Outlook, SharePoint, OneDrive, and Teams, baked into Researcher, Copilot Studio, and Agent Mode in Excel, Word, and PowerPoint. The procurement conversation about whether your firm can use Anthropic tools just ended.
Claude Cowork also moved out of research preview into full general availability. Worth naming: the security concern some have raised about Cowork was never about the model — Cowork runs on the same Claude as every other plan, with the same security profile. The real question is whether you're comfortable letting it browse the web or act on your behalf. That's a permission decision, not model risk.
OpenAI's response: a new $100 ChatGPT Pro tier between Plus and Pro. OpenAI is reacting to Anthropic now, not leading. That's new. Build accordingly.
Quick hits
Tech layoffs hit 78,557 year-to-date. 47.9% attributed directly to AI and workflow automation. Oracle alone cut 20,000–30,000 roles in a single announcement. Tech is leading this wave, not following.
OpenAI bought TBPN. The daily tech talk show is OpenAI's first media acquisition, reportedly for "low hundreds of millions." When AI companies start buying the outlets that cover them, the question of who frames the story gets quietly answered.
Meta launched Muse Spark. The first closed-weight model out of Meta's Superintelligence Labs. If you've been assuming open-source AI is a permanent escape hatch, the door just narrowed.
Anthropic signed a multi-year CoreWeave compute deal. Layered on top of this week's Google TPU and Broadcom commitments. Assume pricing volatility, not pricing decay, on Claude-based workflows through 2026.
The walls just started to move
Every story this week is about the same thing. The wall between vendor and tool (Digits API, M365 × Claude). The wall between firm and client (Perplexity). The wall between audit firm and platform provider (EY). The wall between the CAS practice and the capital that wants to buy it (Modus). All four moved visibly, in the same five days, while the rest of the world was looking elsewhere.
The practices still standing in 12 months will be the ones doing the uncomfortable work of redesigning around what AI makes possible — not the ones bolting it onto what they already have.
What are you redesigning first? If you're not sure where to start, take the free AI Readiness Scorecard — 25 questions, 5 minutes, and you'll know exactly where your practice stands.