Your Meetings Are the Source of Your Tasks

HAKE’s Meeting Agent Proposes. The Human Disposes.

Most task systems start from the inbox. Whatever shouts loudest gets planned first, while the things you actually agreed to on client calls — the scope changes, deliverables and “I’ll get that to you by Friday” commitments — are reconstructed from memory, if at all. questions that separate one you can depend on from one that merely demos well.

We built our week the other way around.

At HAKE Digital, a Claude-based Meeting Agent now sits between our client calls and everything that follows them: preparation briefs before calls, draft documents after them and a daily plan built around what was actually agreed.

The inbox has been demoted to a secondary scan.

This is not a product, app or bot. It is a repeatable Claude workflow assembled from off-the-shelf connectors and plain-English standing instructions.

What It Actually Is

The Meeting Agent is Claude, running in Cowork, connected to the tools we already use:

On top of those connectors sit our standing instructions — house rules written in plain English — and two scheduled tasks.

Each Zoom call’s recording, AI summary and action items are pulled in and combined with calendar, CRM, email and Slack context.

The outputs include:

The Workflow, End to End

Before a Call: Preparation

Claude reads the calendar, previous meeting summaries for the client and the relevant HubSpot records.

It goes beyond surface-level context.

Ahead of a data-migration call for an education-sector client, the agent verified that a specific custom property existed in HubSpot before the call took place. This meant the conversation could begin with facts rather than assumptions.

The output is a preparation brief saved to the working drive, with clear markers wherever the transcripts or available records cannot confirm the scope.

After a Call: Capture and Documentation

A typical instruction from one session was:

“Pull the [client] meeting from 19-06 and prepare the internal and client-facing documents.”

Claude finds the recording, extracts the summary and action items, asks scoping questions where something is genuinely unclear and drafts the required documents.

One call for the same education client produced four branded Word documents in a single run:

• Internal Schedule of Works
• Client-facing Meeting Recap and Next Steps
• Scope Confirmation
• Follow-up Note

All four documents were marked v0.1 — Draft for Review.

They also included [TODO] and [ASSUMPTION] markers wherever information was missing, with no unsupported outcome guarantees.

Continuously: Workload Management

The workload manager reads the Meeting Agent first. Two scheduled tasks run automatically.

Daily Brief

At 08:00, the daily brief begins with the instruction:

“Read the Meeting Agent first.”

It pulls the latest meeting summaries and action items as the primary source of work, extracts only the actions assigned to the correct owner and then carries out a quick Gmail and Slack scan for anything that never reached a meeting.

It then:

• Reconciles tasks against the calendar
• Scores work by impact and effort
• Categorises items as quick wins, big rocks, fill-ins or question marks
• Posts a summary to Slack
• Proposes calendar blocks

Those calendar blocks are only proposals. They are approved, edited or rejected in chat before anything is added to the calendar.

Weekly Review

A weekly review runs on Friday afternoons. It performs a broader reassessment of the workload, including:

• A full priority re-score
• Proposed focus blocks for the following week
• A running backlog file for continuity

 In one week, this materially changed how we planned our work. 

Re-running the review using the Meeting Agent’s summaries revealed that the real workload consisted of three major build and contract items agreed during calls. These outweighed everything sitting in the inbox.

It also corrected an earlier brief that had assigned a colleague’s action items to the wrong owner.

The agent flagged its own mistake rather than compounding it.

The Outcomes

The system has helped us create:

The Guardrails Matter More Than the Automation

The tempting story about AI agents is autonomy: messages sent on your behalf and decisions made while you sleep.

Ours is deliberately designed to work differently.

The guardrails were consistent across every run we reviewed.

Nothing Automatically Ships

Every client-facing document is created as a draft for approval.

Every calendar change waits for explicit confirmation in chat.

As one run stated:

“Per house rules, nothing has been sent — both are drafts for your sign-off.” 

No Fabrication

Where a genuine input is missing, the gap is marked as [TODO] or [ASSUMPTION].

It is never filled with something merely plausible-sounding.

Compliance Flags

Anything involving one of our regulated financial-services clients is automatically tagged [COMPLIANCE] for review before it leaves the business.

Financial promotions are tightly governed, and a draft is not a decision.

Client Data Segregation

Each client’s context is kept separate from every other client’s information.

The Agent Proposes. The Human Disposes.n

That division of labour is the design, not a limitation.

What It Looks Like Across Real Engagements

Education-Sector Client

A single data-migration call became four draft documents, alongside preparation briefs for back-to-back calls that included live verification of CRM custom properties.

Help-Desk Client

A weekly work-in-progress call became an internal scope of works, including research into platform limitations, and a client-facing recap.

Both were produced as drafts.

Partner Agency 

A partnership call became structured notes, actions organised by owner and a draft commercial agreement explicitly marked as not for signature pending solicitor review.

Internal Workload Management

In one daily run, the workload manager extracted more than 25 owner-specific action items from ten meetings and reconciled them against the calendar before proposing the day’s plan.

Built, Not Bought

Perhaps the most useful takeaway is that none of this required a development team. 

HAKE Digital staff assembled a working agent using connectors and standing instructions written in plain English.

The pattern is straightforward:

• Meetings become the source of truth
• Humans remain in the loop by design
• Guardrails are encoded as house rules
• The agent follows those rules consistently 

This is the same pattern we now design and implement for clients through our RevOps and AI advisory work.

The system is designed to ensure that what was agreed during a call is what gets planned, drafted and completed.

It does not promise that you will never miss an action item. Honestly, nothing can.

But it moves the starting point of your week away from the loudest email and towards the most recent agreement.

That changes a great deal.