See what your automations are doing, and approve what matters.
When work runs across several systems, or when an AI agent acts for you, someone needs one screen that shows what’s waiting, what ran, what failed and what needs a decision. I build that screen, connected to the systems doing the work.
It’s a short form: what you need, a few lines about the project, and your email. I read every inquiry myself and reply by email, usually within 24 hours.
A person decides here.
What I build
- Workflow view. What’s queued, running, done and failed, pulled from the systems that do the work.
- Approval queue. Items waiting for a person, with the context needed to decide, and buttons to approve, edit or reject.
- Roles and permissions. Who can see, approve and change what.
- Errors and exceptions. A clear state when an integration fails or data is missing, and a way to retry or hand the case to a person.
- Activity log. What happened, who approved it and when.
- Integrations. Connections to your CRM, store, support desk or database.
Example: refund requests
An agent reads a customer’s email, finds the order and drafts a reply with a refund.
The approval screen shows the email, the order, the draft and the refund amount side by side.
Refunds under a limit you set go out after one click. Larger ones go to a manager.
Anyone with access can pause the agent, and every action appears in the log.
The screen, and the logic behind it
This page is about the interface. The logic that decides and acts, whether a simple automation or an AI agent, is agentic workflow development. One inquiry can cover both.
The screen can only show and do what the connected systems allow. Permissions and access are agreed per project, and I check them before the build. I don’t certify security or compliance.
Why me for this
I build the dashboard and the integrations behind it myself, so the screen and the systems it controls are designed together.
How we’d work together
I usually work for a fixed weekly rate. The work is scoped around a goal rather than a set number of days, and we agree the scope and terms for each project. They’re open to discussion. This runs like any software project: discovery, build and support after launch are agreed separately.
Questions
Is this just a custom web app?
Yes. It’s software development aimed at one job: showing the state of a workflow and putting decisions in front of the right person.
Is this only for AI agents?
No. Ordinary automations and integrations need the same visibility. When an AI agent is involved, the approval steps matter more.
Can it work with the tools we already use?
If they have an API or a database I can access, usually yes. I check that before we agree scope.
Further reading
Event-Driven Growth Systems: Real-Time Personalization and Experimentation for Intelligent Platforms
An event driven personalization platform ties real-time behavioral signals to decision logic and experimentation controls. This guide shows how to log assignment and exposure so a personalization “win” survives replay, with a runnable query that shows how latency can fake a lift.
Articles and guides about this work. They explain methods and trade-offs; they aren’t case studies.
Which decisions should a person make?
Describe the process, the tools it runs through, and the points where someone needs to approve or check something. I’ll suggest what the screen needs to show.