All micro-internships
Micro-internship brief ≈16h across 2 daysbuilder

Build an AI agent that finishes a multi-step task

Agentic AI is the biggest hiring trend of 2026 — systems that plan, use tools, and complete tasks with minimal supervision. Build an agent that takes one instruction and completes a task requiring at least 3 steps and at least 1 tool.

AI Application DevelopmentPrompt EngineeringAPI Integration

Posted by The AI Internships

The work brief

  1. 01Pick a task with real steps: "research today’s top 3 news on X and draft a summary email", "check weather + suggest what to wear + set a reminder", "read a folder of files and produce a report".
  2. 02Build it: code with an agent framework (LangChain agents, CrewAI, OpenAI/Gemini function calling) or wire it in n8n with an AI node + tools.
  3. 03Give the agent at least one real tool: web search, a spreadsheet, calendar, file system, or an API.
  4. 04Run it 5 times. Keep a failure log — agents fail in interesting ways, and hiding that fails this mission.

What you’ll produce

4 deliverables

Submission standard

Submit the task, the repo (or exported workflow), a demo video, and your 5-run failure log. A perfect log with zero failures reads as an untested agent. Include your AI workflow: tools, best prompts, and what you changed from the AI’s output. Your repo's commits should show the fixes you describe in the failure log, and a zero-failure log will be read as untested.

  • The task your agent completes, and its steps

    Short answerRequired

  • Public repo (code or exported workflow JSON)

    Public repo (code or exported workflow JSON) — commit history should show tool wiring and prompt fixes across multiple commits, not one paste. A single dump commit fails the authenticity gate.

    Public linkRequired

  • Public demo video of a full run

    Public linkRequired

  • Your 5-run log: what worked, what failed, what you changed

    Written responseRequired

You’ll complete these inside your private workspace.

How your work is evaluated

The passing benchmark is 70/100.

Genuinely agentic

29%

Multi-step task with real tool use — not a single prompt-response.

Reliability honesty

29%

The 5-run log shows real testing with real failures and fixes.

Implementation

14%

The repo/workflow shows the actual agent wiring.

Really run, really iterated

29%

A flawless zero-failure log, a dead demo link, or a single-commit repo all fail this: real agents fail across real runs and the commit trail shows the fixes.

How we grade your AI usage

30% of your score

Using AI is the point — it’s the skill this certificate proves. You’ll answer three short questions about how you used it: what you asked, what was wrong with its first answer, and what you changed. Specific, honest answers score high. “I pasted the brief and submitted the answer” scores near zero.