Hello fellow keepers of numbers,

Welcome to Friday the 13th. Indulge me with my short story:

The sun is peeking through the vibrant green leaves of the oak tree as you stare out the office window. It’s Friday. You’ve been waiting for this all week. Your friend messaged you the other day - happy hour on him this Friday. It’s been a long week. The clock hits 3pm. It’s finally time. You grab your bag and jacket and start to head out. You feel your phone buzz. Stupidly, you take it out. It’s a message from your most important client. It says “quick question.”

Friday the 13th strikes again.

That was written by yours truly. After I wrote it, I came to the realization that my AI is funnier than me now, which is just really, really annoying. Fun fact: we have another Friday the 13th next month. Maybe, if you’re lucky, I’ll have another short horror story for you next month. Back to regularly scheduled programming…

This week, the bookkeeping world laid to rest Botkeeper, Goldman makes a bet that Fieldguide won’t pull a Botkeeper, and Claude Cowork finally makes its way to Windows. Plus, stay tuned for another Claude Cowork demo and some interesting AI videos from the new Seedance 2.0.

THE LATEST

Botkeeper, one of the earliest AI-driven bookkeeping automation platforms for accountants and firms, announced it is closing and beginning an “orderly wind-down” after 11 years in business. Founder and CEO Enrico Palmerino cited a “perfect storm” of macroeconomic shifts and rapid industry consolidation among Botkeeper’s largest clients, which hit revenue and left the company without a sustainable path forward.

In his letter, Palmerino said Botkeeper’s Infinite platform could clean up years of messy data in minutes, autonomously reconcile accounts, and code more than 80% of transactions with about 98% accuracy, and that the company was two months away from launching a voice assistant (“Cassie”) and autonomous check-scanning. Separate reporting notes the company had raised nearly $90 million in venture funding, including a $25 million Series B in 2020 and a $42 million Series C in 2021, before deciding to wind down.

For current customers, the immediate issue is data and continuity. Botkeeper has told users that client portals will remain open for now and is advising firms to export their general ledger data, transaction history, reports, client lists, and related documentation by February 20, 2026 to ensure they retain complete records for all Botkeeper-served clients. Firms relying on Botkeeper for day-to-day bookkeeping and reconciliations will need to migrate work to staff or alternative systems while contracts, workflows, and deadlines are reviewed.

Why it’s important for us:

There appear to be people on two ends of the spectrum when it comes to opinions about this news. One side suggests this is a tragedy, and they were “before their time.” The other side suggests they were frauds who were really using offshore labor to do what they implied was done by AI.

I don’t know enough about Botkeeper’s situation to make a bold claim on either side. But I can still dissect the news here. The founder is saying they were just a few months away from launching a voice assistant and autonomous check-scanning. Important enough to him that he mentions it in the shutdown letter. Yet they couldn’t even manage to find a buyer and had to outright shut down?

The math isn’t mathing. Seems like potential buyers didn’t see any value in the company or the new products mentioned.

I’m also not sure what the founder means by “rapid market shifts” and “unexpected industry consolidation.” My translation of the shutdown letter: Other applications are now better at doing what we do, and our company is worth negative dollars.

All that said, sometimes companies with a good vision fail. And there has been a lot of support for Botkeeper over the last several days. At the end of the day, it’s a bit sad that they shut down. They had a good vision for the industry. But there are really great options out there for automated bookkeeping now. I think the firms that used Botkeeper will be pleasantly surprised once they get through the mess of converting everything.

Fieldguide announced a $75 million Series C round led by Growth Equity at Goldman Sachs Alternatives to scale its “agentic AI” platform for audit and advisory. The raise brings total funding to $125 million and values the company at about $700 million. Fieldguide’s pitch is that its AI agents work alongside human auditors to execute high-volume, repeatable, data-heavy tasks throughout an engagement.

The company is targeting the audit capacity crunch as its core problem. It cites a current shortfall of roughly 125 million hours of audit capacity, or over $25 billion in unmet demand, with that gap potentially growing to about 600 million hours and $230 billion annually by 2030 if nothing changes. At the same time, CPA exam candidates are at a 17-year low and up to 75% of today’s CPAs are expected to retire over the next decade.

In practice, the platform's agents handle first-pass work like reviewing evidence, checking consistency in testing, and flagging exceptions, while staff focus on judgment, review, and client conversations. Fieldguide says every automated step is logged, attributable, and overseen by a practitioner. The platform is already in use at roughly half of the Top 100 U.S. firms, including some Big Four members, and the new funding is earmarked to expand agent capabilities across more engagement types.

Why it’s important for us:

Fieldguide is one of the leaders in the AI space for audit and assurance. If we’ve learned nothing else over the past 3 years, it’s that it’s apparently very expensive to start and run an AI company.

Fieldguide touts that they’re used by half of the Top 100 U.S. firms. When the Big Four brag about how they have hundreds of AI agents deployed, firms like Fieldguide are what help make that statement (partially) factual.

How do small firms fit into this picture?

The politically correct answer: time will tell.

The opinionated answer you probably came here to read: I’m not convinced that Fieldguide’s differentiators justify the price for the small firms. A combination of custom GPTs, agent skills, and new agentic tools (like Claude Cowork and Codex) might allow small firms to spin up very custom, specific agents quickly and cheaply.

There’s obvious value for an all-in-one platform, especially for firms operating a large A&A practice. To be fair to Fieldguide, I’m not sure their target market is necessarily the small firms. But the same argument can be made for any other similar applications being offered to small firms.

I think all firms should be very selective about the applications they’re choosing to purchase and implement right now. One free plugin from Anthropic for Claude Cowork could make an entire software company worth hundreds of millions completely irrelevant. It’s important to have someone on your team who’s well-informed so you’re not upended by this massive shift. Or even worse, so you don’t spend years paying thousands of dollars for an application that might do the same thing as a $20/mo basic AI subscription.

Anthropic expanded its Claude Cowork research preview to Windows, bringing its desktop agent from macOS to a much larger user base. Cowork runs inside the Claude desktop app and lets you point Claude at a sandboxed folder on your machine where it can read, edit, and create files to complete multi-step tasks like organizing downloads, turning receipt screenshots into expense spreadsheets, or drafting reports from scattered notes.

Under the hood, Cowork uses multiple sub-agents to break complex work into parallel steps, runs in an isolated virtual machine on your computer, and shows progress indicators while it works. You choose which folders it can access, and it can ask for explicit permission before making changes. Cowork can also tap into Claude connectors and the Claude in Chrome browser extension.

Why it’s important for us:

Windows users rejoice! You can safely put the wallet back in your pocket. We all know that, between Claude Cowork previously being locked into macOS and everyone talking about buying Mac minis to run their OpenClaw, you were about to shell out some cash.

Claude Cowork is one of my favorite releases in a very long time. If you’re already a Claude Code user (or other similar tools like Cursor, Antigravity, etc.), this probably isn’t an Earth-shattering release. But the very large majority of accountants are not using those tools. Claude Cowork brings their power into a very simple interface.

In last week’s newsletter, I did a demo of Claude Cowork’s finance plugin, which used some bank and credit card statements to create a chart of accounts, GL, and trial balance from scratch. It also took those numbers and prepared an income statement.

This week, I have a demo of Claude Cowork using the Claude in Chrome extension to log my CPE with the state of Texas.

There are a near infinite number of other applications for Claude Cowork. Any time you’re opening the file explorer to grab a file, ask yourself: could I link this folder and send a prompt to Claude Cowork to do this for me? The answer is almost always yes.

PUT IT TO WORK

I put Claude Cowork to the test to automate my CPE reporting. I let Cowork use the Claude in Chrome tool to take over my browser and put my information into the website for Texas licensees.

Fair warning: there’s a good bit of watching/waiting for Claude to do things, so I’d probably recommend watching on 2x speed.

CPE Reporting with Claude Cowork - Watch Video

WEEKLY RANDOM

ByteDance dropped Seedance 2.0, an AI video model, this week. Genuinely one of the most mind-blowing AI releases I can remember. The videos are incredibly high quality.

There’s one other important thing…

This is a Chinese model, which means there are no IP restrictions, and they may never come.

Several other big AI video model releases, such as Google’s Veo 3 and OpenAI’s Sora 2, initially were very lax with IP restrictions. Because they’re U.S. models, those concerns were addressed within a few days, at most. I’m not sure we’ll ever see this come for Seedance 2.0.

On one hand, it’s kind of awesome because we’ll get some fun videos. On the other hand, I’m not sure we, as a society, are ready for this yet. Will we roll over and just let people create AI-generated videos of famous actors and real humans? Will we only punish people if they’re negligently using the model for deepfakes? Will we ban the model entirely in the U.S.? Who enforces that and how?

Lots of questions to be answered in a very short time. But hey, that’s not a problem for us. So let’s just enjoy some of the videos.