Working with AI APIs — structured, practical, applied
A workshop platform built around doing rather than watching. Each module is a set of real tasks using live APIs — OpenAI, Anthropic, Google Gemini — and the feedback loop is built into every step.
Conditions that make this work
This platform works well under specific conditions. Participants who arrive with a working knowledge of at least one programming language — Python, JavaScript, or similar — move through the material without friction. The exercises assume you can read an error message and know where to look next.
Time matters too. Most participants dedicate 6–9 hours per week. Those who treat it as occasional reading tend to stall around week 3.
Not deep expertise — but familiarity with loops, functions, and HTTP requests. The first exercise involves a live API call within the first 20 minutes.
Workshops are asynchronous, but the assignment rhythm is weekly. Falling behind by 2 weeks makes catching up noticeably harder.
Participants with a concrete use case — automating a workflow, building a tool, integrating AI into an existing product — apply the material faster.
Exercises use real endpoints. Free-tier access to at least one major AI provider is enough to complete the foundational modules.
What the approach actually does
Tasks are built around output, not explanation
Each module ends with something you can run, test, or share. Explanatory text is short — the bulk of time is spent in the editor, not reading slides. Assignments specify the expected output format so you know when you're done.
Errors are part of the curriculum
Rate limit errors, malformed JSON responses, token count overflows — these appear in exercises deliberately. The curriculum includes 7 dedicated debugging scenarios drawn from real API integration failures reported by developers in 2023 and 2024.
Coverage across 4 major AI API providers
OpenAI, Anthropic, Google Gemini, and Mistral are all covered with provider-specific exercises. The final module compares response quality and cost across providers for an identical prompt — using your own API keys and real billing data.
The people around you here
The participant group at any given time includes developers, product managers, and technical founders — roughly 60% are working on something active, not exploring speculatively. Peer review is built into 4 of the 14 workshops.
Vesna Ostrowski
Backend developer, Poland
"The debugging exercises were the most useful part. I'd been avoiding error handling in my API calls for months — seeing it treated as normal curriculum rather than an edge case changed how I structure integrations."
— Vesna Ostrowski, completed Modules 1–6"I came in with a specific project — a document summarisation tool — and left with it mostly built. The peer review in module 4 caught a token management issue I'd missed for 3 days."
— Tarquin Mbeki, product lead, South AfricaCurrent field conditions
The material is updated quarterly. The last revision in Q3 2024 added 3 new exercises covering structured output formats following OpenAI's August 2024 API changes. Provider pricing tables are refreshed monthly.
-
Anthropic Claude 3.5 coverage added September 2024
New exercises reflect the updated context window and tool use API changes introduced in the 3.5 release.
-
Real billing data in Module 12
Participants run identical prompts across providers and compare actual token costs — figures from the previous cohort averaged AU$1.40 per participant for the full comparison exercise.
-
Rate limit scenarios from 2024 production incidents
7 debugging scenarios are drawn from documented outages and API behaviour changes reported between January and October 2024.
Updated Q3 2024
Who belongs here — and who probably does not
What it feels like to move through this
The first week is orientation — getting your environment set up, making a first API call, understanding the assignment format. Most participants describe it as slower than expected, which is deliberate. The scaffolding at the start reduces friction later.
By week 4, the pace shifts. Assignments reference earlier work, and the feedback from peers starts to carry weight. The mid-point — around module 7 of 14 — is where participants typically either accelerate or stall. Those who stall usually do so because of time, not difficulty.
The final module is unstructured by design. You bring a problem, apply the tools from the preceding 13 modules, and submit whatever you built. There is no single correct output. Instructor feedback on final submissions averages 48 hours turnaround.
Environment, first API call, assignment format. Slower pace, higher scaffolding.
Real API calls, error handling, provider comparison. Weekly rhythm establishes momentum.
4 modules include peer review. Feedback loop tightens. Pace increases for most participants.
Unstructured. Bring a real problem. Instructor feedback within 48 hours of submission.