Your CRM needs a write policy

For most of my career, the CRM problem was that nobody updated it. That problem is going away. The new one is that everything updates it, and nobody owns the result.

Once agents write to your CRM, write access becomes a budget decision. Every field that moves money (who owns the account, what stage the deal is in, where the lead came from) needs one named human owner. "The AI updated it" is not an owner.

What changed this month

On 16 September, at its Unbound26 event, HubSpot rebuilt its platform around agents. A redesigned Breeze Assistant now takes an objective and picks the specialised agents to carry it out. A self-updating Smart CRM captures calls, emails and meetings without anyone typing them in. A new Context Home scores how complete that information is. An updated Prospecting Agent monitors more than 40 buying signals and drafts outreach, and Deal Progression suggests CRM updates from call transcripts, with user approval. Leads generated in ChatGPT Ads now flow into the CRM for follow-up and attribution, with Microsoft Advertising being added.

The product is HubSpot's. The question is not. Any team that switches on several of these features one trial at a time, without a meeting about who may change what, ends up with more writers on the same record than it has ever had.

How a record goes wrong while everyone does their job

The mechanism is simple. Take one field: lifecycle stage. A form-fill sets it. The prospecting agent reads a buying signal and moves it forward. A rep on a call hears "we are not ready until Q2", and the notetaker's summary suggests moving it back. An enrichment sync overwrites the company size that the scoring rule depends on. Each write is reasonable on its own. Together they produce a record that says whatever the last writer believed.

Nobody notices, because nothing broke. The dashboard still fills in. The completeness score may even rise, because a full field is not the same as a correct one. What changes is quieter. Sales stops trusting the stage. Reps keep their own notes on the side. The forecast call turns into an argument about whose version of the account is real. And when someone asks marketing why pipeline moved, nobody can say which writer moved it.

Manual CRMs failed by omission. Agent-written CRMs fail by overwrite. The second is harder to catch, because the record looks healthy.

Leads that arrive from several ad platforms at once, each with its own lead form and its own sync, make this worse before a single agent is switched on. The number of writers grows quietly, one integration at a time.

A write policy fits on one page

The answer is not to switch the agents off. For a small team, automatic capture beats manual entry that never happened. The answer is to treat write access the way you already treat budget: someone signs for it.

Three rules I would put in place before the next agent goes live. (1) List every writer: each agent, integration, sync and workflow allowed to change a record. If that list takes more than an afternoon to build, the list itself is the finding. (2) Name one owner for each field that feeds forecast or attribution: lifecycle stage, account owner, deal stage, lead source. A person, not a team, and they decide which writers may touch it. (3) On those fields, agents propose; they do not overwrite. Suggest-with-approval already exists in the tools, and Deal Progression is described exactly that way. Make it the default for the fields that matter, and read a weekly list of what changed and who changed it.

Ownership of CRM data is already a live nerve. In July, HubSpot announced a terms change that would have shared enrichment data across customers by default, then reversed it within days after customer pushback, committing that future enrichment using customer data would be opt-in. Customers reacted fast when the question was who touches their data from outside the building. The same question deserves an answer inside it.

Here is the test I would run this week. For every field that feeds your forecast, can you name the person who owns it and every system allowed to change it? If you can, agents will make your CRM faster. If you cannot, they will make it louder, and you will find out in the pipeline review, not on the dashboard.