Not one major model release this week. And it still might be the most consequential week of the year for how your practice runs. Ramp shipped an “AI operating system” built for accounting firms, your practice management layer started growing sockets for whatever AI you bring, and Thrive put $1 billion on the table to buy firms like yours. And while the vendors were fighting over your methodology, 69% of your peers told a surveyor the hourly billing model is on its way out.
Ramp just shipped the firm operating system — from the spend-management aisle
On Wednesday, Ramp launched Stack and labeled it an “AI operating system built specifically for accounting firms.” In plain terms: a platform where AI agents — AI that doesn’t just answer questions but carries out a sequence of work on its own — run reconciliations, recurring schedules, journal entries, and variance analysis end to end, posting to QuickBooks only after a human approves. You teach Stack your processes and it captures them as skills: editable documents encoding how your firm does the work, like an SOP your agent actually follows. The distribution muscle is real — Ramp partners with 4,500+ accounting firms, 92 of the top 100 CPA firms have clients on the platform, and Stack is free through August.
Credit first. This is a strong start and exactly what we should be demanding from vendors — a product that does the work rather than chatting about it, logs every decision for review, keeps a human gate on anything touching the GL, and contractually treats your encoded processes as your firm’s IP. The teach-it-once loop — run a task with the agent, save it as a skill, it runs the same way every month after — is the encoding loop I’ve spent six months telling you to build, shipped as an onboarding flow.
Now the caution, because almost every number you’ve read about Stack is Ramp-reported. The headline benchmark — Stack at 65.8% versus ChatGPT 5.5, Claude, and Gemini across 200+ accounting tasks — is vendor-commissioned, compares against general-purpose models running with no skills or memory, and tests Claude Opus 4.7 rather than the 4.8 release that shipped six days before launch. What it honestly shows is that the vertical context layer wins. That’s the context-engineering thesis with a vendor’s logo on it — not proof that Stack beats a well-built firm-side setup.
And notice who built it. I’d have expected practice management vendors to get here first — they hold the workflow. Ramp got there instead because it holds something better: a write path to the books and data from over 1 million closes. But that lens is also the limit — Stack sees the close, not your practice. There’s no Xero, no practice management integration, no client communications, and no tax or advisory yet.
The bigger structural problem: there’s no way to add your own connectors or point your own AI at it. No open standard, no external API — a walled garden with excellent landscaping. Even the “your skills are your IP” promise is contractual, not architectural: no source says whether you can export your skills if you leave. Ask that question before you teach any platform your methodology.
The same week, practice management grew sockets
Contrast Wednesday’s other launch. At Karbon Next in San Diego, Karbon unveiled Kai — an “AI coworker” inside the platform, drawing context from 80+ integrations, now in early access. But the strategically interesting announcement wasn’t Kai. It was the public MCP server alongside it. MCP — Model Context Protocol — is the open standard that lets you plug the AI you choose into a vendor’s data: a universal socket instead of a custom cable for every tool pairing.
Karbon isn’t alone. Firm360 shipped a Claude connector for its practice management platform this week, Abacor followed June 4 with one for its meeting and contact assistant, and Kick opened its entire ledger over MCP just two weeks ago. Four open sockets in a fortnight — all in the layers where lock-in is weakest, while the core ledgers keep building their AI behind closed doors.
This is where I want to see things going. Karbon hedging Kai with an open MCP server is the tell: practice management vendors are betting they win by being the context source for whatever AI your firm brings, not by forcing their own assistant on you. For your AI Champion, this is immediately actionable — a firm running Karbon and Kick can now orchestrate work-item status, client context, and ledger work from a single Claude session, on vendor-supported rails. The cross-system agent I’ve told you to be cautious about just got materially safer to build.
Put the two postures side by side, because this is the strategic choice forming under your stack: rent the loop inside a vendor’s walls, or run it on open rails. We won’t be running entire practices from a chat prompt — but in an agentic world, open walls make far more sense than closed ones.
Thrive’s tax agent now has a shopping list — $1 billion of firms like yours
On Friday I wrote that you can’t build Thrive’s tax agent, but you can build the loop it runs on. This week Thrive answered the question that piece left open: what do you do with a 97%-accuracy tax agent? You buy the firms. Thrive Holdings told Forbes it plans to spend $1 billion acquiring local US accounting firms, on top of the $500 million already invested in Crete Professionals Alliance — and the same day, Crete rebranded as Current, promising to bring “Fortune 500-caliber advisory services to Main Street businesses.”
Read it as the encoding loop’s exit strategy. Every acquired firm’s corrections and workflows feed an agent stack the holding company owns, which raises the value of the next acquisition — the compounding flywheel running on M&A instead of organic adoption. The competitive frame just changed: the firm down the street may soon be a Current branch with a self-improving tax agent and a balance sheet behind it. “We’re a local firm competing with other local firms” is an expiring assumption.
The counter-read matters too. Roll-ups have a long history of underdelivering in professional services — culture, client attrition, partner flight — and an AI thesis doesn’t repeal those dynamics. But look at what an acquirer is actually paying for: your client relationships, your encoded methodology, and your QC record. Those are the same three things Friday’s piece told you to operationalize. Own them deliberately and you can compete with Current — or name your price to it.
The profession just admitted the billing model is ending — in a survey its own vendor paid for
The second annual Blue J/CPA.com AI Tax Research Solution Outlook Report landed June 2, surveying 1,000+ US tax professionals. The adoption number first: 60% now use AI for tax research at least weekly, up from 33% a year ago. That’s not gradual improvement — it’s the same step-function pattern we’ve tracked since agentic adoption jumped from 9% to 41% in a single year, now showing up in one of the profession’s most conservative segments.
But the number worth pinning to your wall is this one: 69% of respondents anticipate moving to value-based, hybrid, or fixed-fee billing as AI commodifies routine tax knowledge. That’s the pricing pivot — volunteered by the profession itself, at survey scale, for the first time. Two years ago this argument was a thesis. Now it’s a majority opinion among the people doing the work.
Look at where the time dividend goes, too: of the 84% who say AI saves them time, 50% reinvest it in client response times, 47% in staff work-life balance, and 46% in higher-quality advice. The profession is spending its AI savings on relationships and quality, not headcount cuts — which is what the pivot looks like from the inside.
The honest caveats: Blue J sells AI tax research, so the sponsor benefits from the story the data tells. And Accounting Today’s companion read notes search engines remain prominent — adoption is broad, not yet deep. But 69% of your peers have already concluded the hourly model won’t survive what’s coming. The question they haven’t answered — and you can — is what replaces it at your firm.
Quick hits
Anthropic filed its confidential S-1 on June 1. No financials are public yet, but the model vendor underneath PwC, KPMG, Deloitte, Xero — and probably your own stack — is heading to public markets. Quarterly disclosure will shape model pricing, deprecation schedules, and our first honest look at the economics under your AI bill.
Wolters Kluwer expanded its OpenAI partnership, embedding agentic workflows in CCH Axcess for client data gathering, document classification, and prep steps — with a company-cited 20–30% reduction in manual tasks. The model wars are becoming channel wars: your firm increasingly inherits its model exposure from its tool stack rather than choosing it.
OpenAI pushed Codex beyond developers with role-specific plugins bundling apps, skills, and workflows. Non-developers are about 20% of 5 million+ weekly Codex users and growing 3x faster than developers, and a Corporate Finance plugin is on the roadmap. When it ships, vendor-shipped skills will compete with yours on plumbing — your edge stays context depth.
What this week actually decided
Step back and notice what didn’t happen: no major model release all week, and the roundup didn’t suffer for it. The action has moved up the stack, from the models to the layer that holds your firm’s encoded knowledge. Every story this week is a different answer to the same question — where does your methodology live, and who controls the door?
Ramp says inside our walls, yours by contract. Karbon and Kick say on open rails, bring your own AI. Thrive says sell it to us. And the Blue J survey explains why the question is urgent: 69% of your peers already know the billing model underneath the old way of working is ending. You need an answer of your own, because the vendors clearly have theirs.
There’s a people version of this question too. Every platform that shipped this week assumes your team can brief, direct, and review an agent — and the training most firms will buy to catch up is still a 2024 prompt course. On Wednesday I’ll show you how to grade any AI training before you book it.
I’m at AICPA Engage this week, where most of these vendors will be on the floor — and I’ll be asking them the open-walls question to their faces. Full report from the conference on Friday. In the meantime: where does your firm’s methodology live today — and could you take it with you if you left?
If your honest answer is “I don’t know yet,” that’s the work — and it comes before any tool decision. You need to decide what your practice’s future looks like before you choose the platforms that get you there, and that’s exactly what our AI Practice Transformation program is built to do. The next cohort starts June 26 — four weeks, one three-hour session per week, working through how your firm will operate, price, and compete in this AI world. Details at theaiaccountant.ai/transformation.