Skip to content

Details

Spend a day building a Claude-powered web app that runs tasks, shows live progress, asks for human approval, and gives users a finished document.

Imagine asking an agent to read a set of project files, pull together the requirements, and draft a document you can send to a client. Where does the user submit the task? How do they see which tools the agent has called and what it’s working on? Can they review a proposed file operation before it happens? And if they close the page, can they come back to the conversation and find the finished document?
In this hands-on workshop, you’ll build a Client Deliverable Workspace to answer those questions. Users submit a task and provide sample project materials. The agent reads the files, prepares a document, and asks for approval before saving the final version.
You’ll practise Agent Harness Engineering by building the interface, tools, execution controls, and state management around the agent. You’ll try the complete workflow first, then use a starter codebase to connect the frontend, backend, agent runtime, and session storage. At the end of the day, you’ll run the app and test how it behaves when things go wrong.

## 🛠️ What you’ll build in one day

  • Application Architecture · Connect the app to the agent

Trace a request from the browser through the backend to the agent runtime and its execution environment. Wire up the pieces that turn a user request into an agent run.

  • Streaming & Tool Visibility · Show what’s happening live

Turn runtime events into streamed responses, tool call records, and task status updates. Show users which tools ran, what they returned, and whether the task is still running.

  • Workspace & Human Approval · Put users in control of file operations

Give the agent limited access to project files and require approval for selected operations. Implement approval, rejection, and cancellation, and connect each decision to the running task.

  • Sessions & Deliverables · Reopen past work and view the finished document

Save conversations and execution records so users can reopen a session and view the complete document alongside the task that produced it.

  • Failure Testing & Debugging · See what happens when a run goes wrong

Test tool errors, rejected operations, interrupted runs, and cases where the agent claims it finished without producing a deliverable. Check what the app shows in each case and where to start debugging.

## 🎯 What you’ll take away

  • A working agent web app you can run locally. Submit a task, follow its progress, approve a file operation, and open the finished document.
  • A practical understanding of agent app architecture. Know where the interface, backend, runtime, tools, and session storage fit, and which layer to investigate when something breaks.
  • A project you can keep building on. Take home your code, saved session records, a sample client deliverable, a failure-testing checklist, and architecture notes.

The workshop focuses on a local, single-user app. Production authentication, multi-tenancy, billing, deployment, and reliable recovery from arbitrary interruptions are outside the workshop scope.

## 👥 Who should attend?

This workshop is for developers and AI engineering learners who know some JavaScript or TypeScript and want to turn an agent demo into a web app people can use. Familiarity with HTTP APIs, asynchronous programming, and basic agent tool calling will help you follow the exercises. Frontend experience is helpful.
You can attend this workshop independently of the 15 November Claude Agent Workshop.
Please bring a laptop. Before the workshop, set up a code editor, Git, and the required Node.js environment, download the starter repository, and complete the setup check. The SDK version, runtime access requirements, and any usage costs will be confirmed before registration.

## 💬 Instructor

To be confirmed.

## 💻 Event details

  • Date · Sunday 29 November 2026
  • Time · 10:00 am–4:00 pm Melbourne time (AEDT)
  • Location · Melbourne city centre
  • Format · In-person, hands-on workshop
  • Language · English
  • Early bird · A$328 per person
  • Standard · A$389 per person

Early bird registration closes on 22 November 2026 at 11:59 pm Melbourne time (AEDT).

Related topics

Events in Melbourne City, AU
Artificial Intelligence
Artificial Intelligence Programming
Job Search
Cloud Computing
Education & Technology

You may also like