AI agents are here. Is your firm ready for them?

Every partner conversation about AI eventually gets to the same question: “when do we start?” Our answer is almost always the same, and almost always unwelcome: “you started three years ago, and you didn’t know it.”

AI agents — the kind that can actually read your engagement files, draft your client emails, and chase your outstanding queries — are not a thing you turn on. They are a thing you feed. And what they eat is your firm’s data.

Which means the real question is not “is AI ready for my firm?” It is “is my firm’s data ready for AI?”

What an AI agent actually needs to be useful

Let us be specific about what separates a useful agent from a glorified chatbot. A useful agent can:

• Tell you which clients have outstanding queries older than 14 days.
• Draft a follow-up for each of them, citing the specific documents outstanding.
• Spot that a client replied to your request last Thursday and mark the query as resolved.
• Pull the right engagement letter, find the right clause, and quote it back to you.

None of those tasks are hard for the AI. They are hard for the AI if your data is a mess. Because every single one depends on the agent being able to answer a prior question: which engagement? which client? which document? which query?

If those questions have ambiguous answers in your systems, the agent cannot do its job. It doesn’t matter how good the model is.

The data problems that quietly kill AI projects

In conversations with firms, we see the same half-dozen data problems over and over. None of them are dramatic. All of them are blockers.

1. The same client, three times over. “Smith Family”, “Smith Family Trust”, and “The Smith’s” are three contacts, three engagement histories, three sets of documents. A human notices. An agent doesn’t. Until you merge, the agent is working with a third of the context on every interaction.

2. Documents without homes. The CT600 for FY24 is in someone’s Outlook. The engagement letter is on the shared drive. The signed copy is in a DocuSign folder nobody has opened in eight months. The agent cannot cite what it cannot find.

3. Queries in your head. “I asked David about the VAT invoices on a call three weeks ago” is not a query. It is a memory. An agent can track open queries, draft follow-ups for them, and close them when resolved — but only if they exist in a system somewhere.

4. Engagements without structure. “Tax work for Jones & Co” is one line in a spreadsheet. “Corporation Tax Return FY25, due 31 Jan, assigned to Sarah, awaiting bank statements, blocked” is a workable engagement. The agent’s usefulness is a direct function of how much of that structure exists.

5. Contact data that lies. Sarah’s new email address came in six months ago. Her old one is still on the engagement letter, the invoicing system, and three distribution lists. When the agent drafts a reminder, which address does it use?

6. Status fields nobody fills in. An engagement marked “in progress” for 18 months is not in progress. It is either done and un-closed, or abandoned and un-archived. The agent can only work with what the fields say.

The good news: this is the same cleanup you needed anyway

Every single one of those problems was costing your firm real money before AI showed up. Duplicate clients cost you billable time. Lost documents cost you deadlines. Un-tracked queries cost you write-offs. AI is just the first tool that makes the cost visible and immediate.

So the honest framing of “getting ready for AI” is: do the data hygiene work you have been putting off for a decade. The AI is not a new reason to do it. It is finally a good reason.

Concretely, the preparation looks like:

1. One client record per client. Merge duplicates. Pick a canonical name. Move on.
2. Every document lives in one system. It doesn’t matter which one, as long as “which one” has an answer.
3. Every query has a record. Written down, attached to an engagement, with a status.
4. Every engagement has a state. “In progress”, “blocked on X”, “awaiting review”, “done”. Not just “open”.
5. Contact data has a single source of truth. Not three systems that almost agree.

That is the groundwork. A firm that does it is ready for AI. A firm that skips it is ready for disappointment.

How Cheku does the cleanup for you

Here is where we will stop pretending we wrote this as a neutral piece.

We built Cheku on the assumption that firms have exactly the data problems above, because every firm we have ever spoken to does. So the product does the cleanup as part of using it.

• Clients, contacts, and engagements live in one place. Not a plugin, not an integration, not a spreadsheet. The agent reads from the same records you edit, which means the data you maintain is the data the AI works from. No sync drift.
• The Agent Inbox surfaces the duplicates. When Cheku notices two contacts that look like the same person, it flags it. You approve the merge or dismiss. The mess gets cleaner every week you use it, without a dedicated cleanup project.
• Queries are first-class. Every query is attached to a client, an engagement, and a status. The agent can chase the ones going stale, because the ones going stale are visible to it.
• Documents are tied to engagements. When you upload a bank statement to Jones & Co’s Q4 reconciliation, the agent knows what it is, which engagement it belongs to, and which query it resolves. Next time you ask “did we get the Q4 bank statements from Jones?”, the answer is instant.
• Fields have meaning. Engagement status, client type, region, jurisdiction — these are structured fields the agent reads, not free text it has to guess at. The schema is the product.

The effect is that adopting Cheku is your data readiness project. You are not doing a six-month cleanup exercise and then picking an AI tool. You are moving your firm into a system where the cleanup is the daily workflow, and the AI is a consequence of it.

The window is now

AI agents are not a future technology. They are here, they are useful, and the firms that adopt them early are going to spend the next three years compounding an advantage that is very difficult to catch up on.

But the adoption curve is not gated by your budget. It is gated by your data. Get that right, and any half-decent agent will pay for itself inside a quarter. Get it wrong, and no amount of AI is going to help.

If you want a shortcut — a platform where the data discipline and the AI live in the same system, designed from day one to work together — that is exactly what Cheku is. And you can start free.