Thirty days. From Agent(model).run_sync("hello") to a script that batches, retries on bad JSON, votes for consistency, and tracks total tokens. Same six prompts you saw on day 1 — rate yourself again.
What's next?
If you've done Automation Beginner, you now have all three primitive layers — Python (language), Composio (acting on the world), LLMs (language models). Combine them.
If not, Automation Beginner is the natural complement: Python that calls real-world services, with the same production patterns (helpers, retry, batch) you just learned for LLMs.
Beyond v1: AI Intermediate adds tool calling (the LLM triggers Python functions), output validation suites, eval frameworks, agent loops. AI Advanced adds embeddings, RAG, model routing, caching, production guardrails. The kit you have here unlocks all of them — they're combinations and refinements of these same primitives.
24 small Python LLM scripts across four weeks plus the integration synthesis. Your kit:
| Capability | Lessons |
|---|---|
| Mental model + calling | L1 (mental model), L2 (Agent), L3 (specificity), L4 (iterating) |
| Four task verbs | L5 (summarise), L6 (infer), L8 (transform), L9 (expand) |
| Few-shot + structured output | L10 (few-shot), L11 (JSON), L12 (parse failures), L13 (schema) |
| Multi-turn + system prompts | L15 (conversation list), L16 (multi-turn), L17 (system prompt) |
| Cost + refusals | L18 (tokens), L19 (refusals) |
| Synthesis | L20 (system + multi-turn + JSON) |
| Production patterns | L22-L28 (batch, helpers, retry-on-bad-output, self-consistency, cost-aware, integration) |
Any LLM-using script you'll read or write from here is a composition of these.
Deferred to AI Intermediate (when v1 expands beyond 3 tracks):
promptfoo, inspect_ai)AI Advanced adds: embeddings, RAG (retrieval-augmented generation), model routing, response caching, prompt versioning, production guardrails (PII detection, content moderation, jailbreak resistance).
None of these are required to write your first dozen useful LLM scripts. You have enough.
→ Automation Beginner if you haven't done it. The full kit — Python primitives + LLM calls + Composio tools — is the v1 north star.
→ Or build something. The kit you have can: classify customer emails, summarise long docs into briefs, extract structured data from messy text, hold a multi-turn chat with persistent memory, route messages by sentiment. Pick a need and try it.
Rate the prompts below as honestly as you did on day 1.
Thirty days. From Agent(model).run_sync("hello") to a script that batches, retries on bad JSON, votes for consistency, and tracks total tokens. Same six prompts you saw on day 1 — rate yourself again.
What's next?
If you've done Automation Beginner, you now have all three primitive layers — Python (language), Composio (acting on the world), LLMs (language models). Combine them.
If not, Automation Beginner is the natural complement: Python that calls real-world services, with the same production patterns (helpers, retry, batch) you just learned for LLMs.
Beyond v1: AI Intermediate adds tool calling (the LLM triggers Python functions), output validation suites, eval frameworks, agent loops. AI Advanced adds embeddings, RAG, model routing, caching, production guardrails. The kit you have here unlocks all of them — they're combinations and refinements of these same primitives.
24 small Python LLM scripts across four weeks plus the integration synthesis. Your kit:
| Capability | Lessons |
|---|---|
| Mental model + calling | L1 (mental model), L2 (Agent), L3 (specificity), L4 (iterating) |
| Four task verbs | L5 (summarise), L6 (infer), L8 (transform), L9 (expand) |
| Few-shot + structured output | L10 (few-shot), L11 (JSON), L12 (parse failures), L13 (schema) |
| Multi-turn + system prompts | L15 (conversation list), L16 (multi-turn), L17 (system prompt) |
| Cost + refusals | L18 (tokens), L19 (refusals) |
| Synthesis | L20 (system + multi-turn + JSON) |
| Production patterns | L22-L28 (batch, helpers, retry-on-bad-output, self-consistency, cost-aware, integration) |
Any LLM-using script you'll read or write from here is a composition of these.
Deferred to AI Intermediate (when v1 expands beyond 3 tracks):
promptfoo, inspect_ai)AI Advanced adds: embeddings, RAG (retrieval-augmented generation), model routing, response caching, prompt versioning, production guardrails (PII detection, content moderation, jailbreak resistance).
None of these are required to write your first dozen useful LLM scripts. You have enough.
→ Automation Beginner if you haven't done it. The full kit — Python primitives + LLM calls + Composio tools — is the v1 north star.
→ Or build something. The kit you have can: classify customer emails, summarise long docs into briefs, extract structured data from messy text, hold a multi-turn chat with persistent memory, route messages by sentiment. Pick a need and try it.
Rate the prompts below as honestly as you did on day 1.
Create a free account to get started. Paid plans unlock all tracks.