AI is coming for professional services. Here's what that actually means.
by Ivor Padilla
Co-Founder & Engineering Director
On January 30, Anthropic released 11 open-source plugins for Claude Cowork covering legal, finance and marketing workflows. By February 5, enterprise software stocks had lost $285 billion in market cap. Thomson Reuters dropped 16%. LegalZoom fell nearly 20%.
The message from investors was clear: AI is no longer adding features to existing software. It is starting to replace the software.
Around the same time, an American AI agency founder's post went viral with a blunt warning to service businesses: if your model is "we do X for clients" and AI can do X, your margins are about to compress. His advice? Build products. Move fast. Ship in days, not months.
He's not wrong. But he's talking about a different world than the one our clients operate in.
The gap between the AI hype cycle and your Monday morning
If you run a law firm, an advisory practice or a consultancy in the EU, here is what your Monday morning actually looks like. A stack of contracts that need clause-by-clause review. Client onboarding documents that need data extracted, validated and filed. Reports that need to be assembled from five different sources, checked for accuracy and delivered before a regulatory deadline.
None of this is glamorous. Most of it is repetitive. All of it requires precision.
The "build a product in a week with AI" crowd is solving a different problem. They are building consumer apps, marketing tools and SaaS products where a 90% accuracy rate is fine because the stakes are low. If a marketing headline is slightly off, you rewrite it. If a contract review misses a liability clause, someone gets sued.
This distinction matters more than anything else in the AI conversation right now. And almost nobody is talking about it.
What Claude Cowork actually changes (and what it doesn't)
Let's be specific about what Anthropic shipped. Cowork plugins can automate workflows across legal review, financial modeling, sales forecasting and marketing campaigns. The plugins use Model Context Protocol (MCP) for API extensions, meaning they can connect to your existing tools.
For generic knowledge work, this is a genuine shift. A marketing team that used to need three tools and a coordinator can now run campaigns from a single AI interface. A finance team can generate models without switching between spreadsheets and analysis platforms.
For regulated professional services, the picture is more nuanced.
What Cowork does well: First-pass document summarization. Drafting standard templates. Research across large document sets. Extracting structured data from unstructured sources.
What it does not solve: Data residency. Your client contracts contain personal data subject to GDPR and the EU AI Act. Where is that data processed? On whose servers? Under which jurisdiction's rules? Cowork runs on Anthropic's infrastructure. For many EU professional services firms, that's a non-starter without additional architecture.
It also does not solve integration with the specific tools your firm already uses. Your practice management system, your document storage, your billing system, the legacy software your IT team has been meaning to replace for three years. Cowork's MCP plugins are a start, but connecting AI to your actual stack requires building the bridges between them.
And it does not solve accuracy at the level professional services demand. A contract review agent that works 95% of the time sounds impressive until you consider that the 5% failure rate means one missed clause in every twenty contracts. In legal work, that's not a rounding error. It's malpractice risk.
The real opportunity is not "replace everything with AI"
The firms that will benefit most from this wave are not the ones that try to replace their teams with Claude Cowork overnight. They are the ones that identify the specific, high-volume workflows where AI adds precision and saves hours.
Three patterns we see working in practice:
Contract review and classification. An agent reads incoming contracts, extracts key terms (renewal dates, liability caps, termination clauses) and flags anomalies against your standard templates. A lawyer still reviews the output. But instead of reading 40 pages, they review a structured summary with flagged sections. The review that took 90 minutes takes 15.
Client onboarding. Documents arrive in different formats from different sources. An agent normalizes them. Extracts required fields. Validates completeness. Routes incomplete submissions back to the client with specific requests. The admin work that consumed a full day per new client drops to an hour of oversight.
Report generation. Data sits in three systems. An agent pulls from each, assembles a draft report in your template, flags inconsistencies between sources. Your team reviews and refines. The two-day assembly process becomes a half-day review process.
In each case, the AI does the repetitive work. The human does the judgment work. The result is not fewer people. It is the same people doing more valuable work.
Why "move fast" looks different in regulated industries
The "ship in a week" advice is sound for startups and agencies. It does not translate directly to firms handling regulated client data.
Here is what "move fast" actually looks like for a professional services firm:
Week 1: Pick one workflow. Map it. Identify where the volume is and where the errors happen. Define what "better" means in measurable terms: time saved, error rate, documents processed.
Week 2: Build a working prototype. Run it against real data in a controlled environment. Measure the results against your baseline.
After the pilot: You have numbers. Not a pitch deck, not a strategy document. Actual before-and-after metrics. Hours saved. Error rates compared. Documents processed per day. That gives your partners the evidence to decide what to automate next, with a clear roadmap for the next 90 days.
This is not slower than the startup approach. It is more disciplined. Because in professional services, the cost of getting it wrong is not a bad product review. It is a lost client or a regulatory violation.
The EU advantage nobody talks about
There is an irony in the current AI conversation. Most of it happens in American English, from American companies, aimed at American markets. The assumption is that EU regulation is a drag on AI adoption.
The opposite is true for professional services.
Your clients chose you partly because you handle their data responsibly. GDPR compliance, EU AI Act readiness, data residency within the EU, audit trails for every decision. These are not obstacles. They are the reason your clients trust you with their most sensitive work.
When you add AI to your operations and you can show clients exactly where their data is processed, that no data is used to train models, that every AI-assisted decision has a human review step and an audit trail. That is not a limitation. That is a selling point.
The firms that figure this out first will not just survive the AI wave. They will use it to deepen the trust that makes their business work.
What to do this week
You don't need to restructure your entire firm. You need to start with one workflow.
- Identify your highest-volume repetitive task. The one where your team spends hours on work that follows a pattern.
- Measure it. How long does it take today? How many errors or rework cycles does it generate?
- Test whether AI can handle the pattern. Not perfectly. Not replacing anyone. Just doing the first 80% so your team can focus on the 20% that requires judgment.
- Measure again. Compare. Decide whether to expand.
That's exactly what a 10-day pilot is for. You pick one process, build the automation, measure results and walk away with a working system plus a roadmap for what to do next.
The $285 billion market reaction is a signal, not a sentence. It means the market believes AI will reshape professional services. The question is whether that reshaping happens to your firm or is directed by your firm.
Start with one workflow. This week.