Sales ops in the chat: an AI agent queries HubSpot and notes every new lead

In a sales team, the CRM is both the source of truth and the thing everyone avoids opening. To answer "which deals are stuck?" or "who owns this company?", you have to navigate filters and views, and every new lead arrives with no context beyond what the form collected. These are two background tasks, useful and repetitive, that get pushed back because they are never urgent. They suit AI agents well: they call for method and sources, not commercial judgment.

The inspiration: the CRM in the conversation

The starting point is a use reported by Hermes Agent users, summarized in a card titled "sales ops in the chat: CRM search, lead notes, email triage". The idea is to take the CRM out of its interface: you ask the question like you would a colleague, in the chat, and the answer arrives with the relevant records. The same use adds automatic lead preparation, with a context note on every new record.

We keep two building blocks, plain-language search and the lead note, and leave aside the email triage of the card, which assumes access to a mailbox. Atako applies the idea with two agents whose access is deliberately limited, and with a simple rule for any change: it is shown before it is made.

What the two agents do

The scenario deploys two agents, both connected to your HubSpot.

The CRM assistant answers questions about contacts, companies and deals in plain language. It starts with a search, opens the records that matter, then answers in a short, structured way: figures or records first, one line of interpretation next, and every record cited is named so you can find it. At startup, it lists the five most advanced deals of the pipeline you named, or of the default pipeline, and summarizes them in the chat with the next action it suggests.

When you ask it to change something, a stage, a property, an owner, a task, it first prepares the change: it shows the record, the current value and the new one, and waits for your confirmation in the chat. A confirmation covers only what was shown. After the operation, it says exactly what it changed.

The lead analyst takes care of new arrivals. On the first run, it processes the contacts created in the last seven days; after that, every morning at 8:00, those created in the last 24 hours. For each one, it reads the record, studies the company on the public web and adds a research note to the record: the company in two lines, the context of the lead, why it matters for your offer and a suggested next action, with its reason. It cites its sources, marks what is uncertain and writes "not found" rather than guess. When the next action needs one, it creates a HubSpot task. Each run ends with a summary in the chat.

What the agents do not do

The limits are part of the design. Neither agent contacts anyone: no email, no message, no invitation. The analyst only adds notes and tasks and changes nothing else on the records. The assistant neither creates nor modifies a record without your confirmation, and never deletes anything: the delete tool is not granted.

When data is missing, empty or ambiguous, for example two records with the same name, the assistant says so and asks instead of guessing. No agent invents a figure, a name, a job title or a quote. As for personal data, the analyst notes only professional, public information. The decision to contact someone stays human, which is the principle of a human-in-the-loop design.

How to set it up in Atako

From the page of this use case, deployment takes four steps.

1. Connect HubSpot. Atako guides you through authorizing access, and the connection is checked before anything is created. If you are not an administrator, the connection stays personal.
2. Provide your context. Describe your offer and your ideal customer in a few sentences, and name the pipeline to follow if you do not use the default one. The agents use it to judge what matters in a lead.
3. Deploy. A single action creates the two agents and the "Sales ops" project. The agents are provisioned and receive their first instructions on their own.
4. First results, then a daily rhythm. The assistant summarizes the five most advanced deals in the chat, and the analyst adds its first notes to the HubSpot records. After that, the analyst schedules its 8:00 run every day on its own.

If you do not have an account yet, creating one starts your trial and the scenario is configured right after. If your administrator has not allowed you to create agents, the deployment tells you and stops there.

Where to see the results

Research notes are read in HubSpot, in the activity of each contact or deal, along with any tasks the analyst created. The assistant's answers and the summary of each analyst run are read in their chats. In the Atako project "Sales ops", you find the how-it-works page and the work tracking, and on the Atako floor, the two agents appear with their current activity.

To query the CRM, write to the assistant in its chat, for example "which deals are stuck at the proposal stage?". Requests are handled in the same conversation, with the same memory of what has already been done.

Guardrails

Access is deliberately narrow. The HubSpot connection is granted to the agents with a limited list of sixteen tools: search and read contacts, companies and deals, update contacts and deals, create notes and tasks, associate records. An agent that sees a short list of tools uses them more reliably than one lost in a catalog of hundreds. Nothing is granted by default, every access can be removed, and the agents reach neither your mailbox nor any other account.

Human validation sits at the decision points. You confirm every change the assistant proposes, you read the notes before contacting anyone, and you decide what to do with each lead.

What to expect

Expect notes to be most useful for companies with a public presence: for a very small business with no website, the analyst will write "not found" rather than invent, and that is the intended behavior. The analyst does not reread its earlier notes in HubSpot: it keeps in its memory the records already processed, and a record created before it started is covered only by the first seven-day run. The value comes from regularity: every new lead arrives with context, and every question about the CRM gets its answer in a few lines instead of several clicks.