Skip to content

Personal agent or custom build?

Your AI can do the work. Who keeps the record?

The short answer

Start with a personal agent for occasional drafts and experiments. When work needs shared records or consequential actions, check your existing tools for enforced approvals, permissions and logs. Build the missing controls around the workflow.

Start with the job. Then check the controls.

Personal AI agents can read your inbox, research a prospect and draft a reply using tools you connect. For some jobs, that is enough.

When the work affects customers, money or your team, ask where the record lives and who controls the next action. You may need better settings, a connection between existing tools, or a custom build.

Two ways to do the work.

Some ready-made agents already support shared access, approval rules and audit logs. A custom build still has to implement those controls correctly. Compare the actual setup, not the label.

A personal AI agent

You connect a ready-made agent to your tools and tell it what you want.

Where it helps

  • A small starting point. Try a draft or research task without commissioning custom software.
  • Flexible instructions. Change the task instructions as you learn how the work should run.
  • Useful first-pass work. Read, summarize, sort and draft, with someone checking the result.
  • A way to test the idea. Find out whether the task is worth repeating before designing a larger system.

What to check

  • Check the record. Can your team find the inputs, drafts and decisions later? Can you export them?
  • Check the permissions. Instructions alone are not an approval control. Look for blocked actions and enforced review steps.
  • Check shared access. Team visibility, audit detail and retention depend on the product, plan and configuration.
  • Check the cost and output. Subscriptions, usage and connections may cost money. Repeated requests can produce different answers.

A custom build

A tool designed around your workflow, with business records and permissions connected to AI.

Where it helps

  • Records your business controls. Keep the authoritative record in agreed accounts, with access and exports defined in writing.
  • Rules in the software. Design permissions and approval steps into the actions the agent can take.
  • A shared history. Give the team a common view of work, decisions, approvals and failures that the system records.
  • Room to change the AI. Separate business records from the model provider. Switching models still needs integration and testing.

What to check

  • Design and build time. The workflow, data and acceptance checks need agreeing before implementation.
  • Changes need an owner. A process change may need a software change, testing and an updated operating rule.
  • Upkeep continues. Hosting, model usage, connections, access reviews and updates need someone responsible.
  • Not every job needs a build. An occasional task or a control your current tools already provide may need no new software.

What happens if the answer is wrong?

If a mistake only means redoing a draft, a personal agent may be enough. If it affects money, trust or a commitment, require the right controls. Check your existing tools first. Build what is missing.

Choose a starting point, then check what your current tools can enforce.
If this is trueStart here
You are testing whether an idea is usefulPersonal agent
A mistake only costs you a redoPersonal agent
It is just you, and the task is occasionalPersonal agent
Prices, availability or commitments depend on the answerSystem controls
The work goes to a customer and needs approvalSystem controls
More than one person needs to see the workShared business records
You may need to show what happened laterA reviewable action history
The data is part of what makes the business valuableAgreed control, retention and export

System controls may already exist in your CRM or agent platform. A custom build or targeted integration makes sense when those tools cannot meet the workflow's requirements.

Give the AI a system to work inside.

Let the AI research and draft. Keep records and permissions in the system responsible for the work. Require a named person to approve consequential actions.

That system may already exist in your business. You do not have to replace every tool or start with a custom build.

  1. Research and draft

    The agent prepares a prospect summary and a reply using approved information.

  2. Record and review

    The system saves the draft against the lead. A person checks the source facts and proposed reply.

  3. Approve the next action

    An enforced approval step gates the send action. The system records the decision and result.

Illustrative workflow, not a client result or a connected service. Start with draft-only access, test the approval boundary and agree who handles missing information or failed connections.

See where the work gets stuck.

Send us your website for a free Mini Teardown. We review your public footprint and identify a likely bottleneck, with sources and any gaps made clear.

It is a starting hypothesis from public information. It does not inspect your private systems or prove which solution your business needs.