Major pathway · Two guided projects
Education
Design useful hints and activities for other learners.
STEMTIQ AI LAB · IDEAS INTO WORKING AGENTS
AI Lab
Explore agents at work. Understand the decisions. Build your own.
Start the first lessonView the learning pathIllustrated lesson previews · Open a lab to run your code.1 / 4
ONE LEARNING PATH · SIX SHORT COURSES
From your first idea
to a tested agent.
Start with the basics. Learn tools, evidence, and evaluation. Then build for a field you care about. Follow 24 bite-sized lessons in order, with real code to run at every step.
Begin the curriculumChecking this browser’s progress…
Already comfortable with functions and conditions? Review the tools course prerequisites.
UnderstandThree small ideas. Visual steps. Check your thinking.
Build & runPredict the result. Edit code. Inspect output and tool calls.
Explain & extendTry a new input. Keep your evidence. Apply the skill.
01Beginner · 4 interactive lessonsUnderstand agents. Write your first code.What is an agent, and what does the code actually do?
Start here when: No coding or AI background required.
Start with a question you can check: did anything actually happen? Separate model responses from program actions, then use JavaScript to turn a small goal into a working rule. No coding background is assumed.
Course overview & build checkpoint- 1.1A model is one part of the systemDistinguish generated text, a fixed workflow, and a tool-using decision loop.Learn · code · experiment
- 1.2Your first few lines of codeRead values, name variables, and write a function with an input and a return value.Learn · code · experiment
- 1.3Make a rule you can testUse arrays, comparisons, conditions, and a bounded loop to inspect choices.Learn · code · experiment
- 1.4Turn an idea into a clear goalWrite explicit success criteria, allowed actions, and a stop condition before implementation.Learn · code · experiment
Your practical outcomeA room-selection function that checks availability, capacity, and budget, with an explicit no-match result.
02Beginner · 4 interactive lessonsBuild the agent's control loop.How does a decision become an observable action?
Start here when: Use functions, objects, conditions, and explicit success checks from phase 1.
A tool request is a proposal. Build the code that validates it, executes an allowed operation, retains the observation, and stops. Then decide which information belongs in temporary state and which may be kept.
Course overview & build checkpoint- 2.1A tool needs a clear contractDefine a tool's purpose, argument schema, returned fields, and permitted effects.Learn · code · experiment
- 2.2Decide → act → observeDispatch an allowed tool request, feed its result back, and finish within a turn limit.Learn · code · experiment
- 2.3What should the agent remember?Separate current task state, an interaction's history, and deliberately retained memory.Learn · code · experiment
- 2.4Return something another program can useParse and validate a result, preserve its evidence fields, and handle an error without inventing success.Learn · code · experiment
Your practical outcomeA bounded controller and a result validator that distinguish missing data, unsupported answers, and failed reads.
03Intermediate · 4 interactive lessonsBuild answers on evidence.Where did the answer come from, and is that source enough?
Start here when: Trace a tool loop and validate its structured inputs and outputs.
Search returns candidates. Learn to inspect them, attach the right source, preserve disagreement, and make missing context visible. Keep instructions inside retrieved text from acquiring the authority to control tools.
Course overview & build checkpoint- 3.1Find a useful piece of informationExplain document chunks, keyword retrieval, embeddings, and the difference between retrieval and generation.Learn · code · experiment
- 3.2Make a claim you can supportConnect each claim to the right source, retain its date, and distinguish contradiction from missing information.Learn · code · experiment
- 3.3Give the agent the right contextSelect relevant instructions, observations, and memory within a limited context budget.Learn · code · experiment
- 3.4Data is not an instructionTreat retrieved text and optional image text as untrusted input; separate policy checks from tool permissions.Learn · code · experiment
Your practical outcomeSource-checking functions that retain evidence IDs, report conflicts and omissions, and enforce a preview-only action boundary.
04Intermediate · 4 interactive lessonsTest, debug, and control the work.What happens when a tool fails, a result is wrong, or an action needs permission?
Start here when: Use sourced evidence, explicit unknowns, and a controlled tool loop.
Useful behavior includes stopping at the right time. Bind approval to a particular action, write checks that can reject a confident answer, handle failed reads, and inspect repeated work in the actual tool trace.
Course overview & build checkpoint- 4.1Ask before changing somethingApply least privilege and check explicit approval at the tool that would cause a side effect.Learn · code · experiment
- 4.2Test behavior, not confidenceBuild representative and held-out cases; distinguish code checks, human review, and fallible model graders.Learn · code · experiment
- 4.3Fail clearly and recover carefullyClassify errors, bound retries, and distinguish a retryable read from a potentially duplicated write.Learn · code · experiment
- 4.4See where time and calls wentCompare correctness, duration, and call counts; use token and cost metrics only when a real runtime reports them.Learn · code · experiment
Your practical outcomeAn approval gate, a behavior checker, a bounded retry policy, and a per-run cache with observable call counts.
05Advanced · 4 interactive lessonsDesign connected systems.Which architecture and tools fit the task—and which complexity can you avoid?
Start here when: Validate tool calls, enforce permissions, inspect traces, and run an evaluation set.
Compare a fixed route, a bounded proposal-and-check loop, and a tool adapter. Make each boundary explicit before exploring the separate local LangGraph project. The browser exercises execute code; the local extension adds an actual model when configured.
Course overview & build checkpoint- 5.1Choose a workflow before adding a teamCompare fixed sequences, routing, parallel independent work, and a model-directed loop.Learn · code · experiment
- 5.2A proposal must earn acceptanceKeep proposal generation separate from independent verification and bounded revision.Learn · code · experiment
- 5.3Connect tools without giving away controlDistinguish local function calls, MCP tool interfaces, and A2A task delegation.Learn · code · experiment
- 5.4Run with real libraries and model limitsIdentify the framework, model, runtime, license, data destination, and limits in an actual SDK project.Learn · code · experiment
Your practical outcomeA routed checker, an independently verified proposal, and a permission-aware adapter; plus a downloadable local SDK project to extend.
06Advanced · 4 interactive lessonsMake a project you can explain.Can you show what your agent does, where it fails, and why its design fits your chosen field?
Start here when: Complete the shared skills; choose a project scope that fits the available environment.
Start from one person's problem. Build a complete small workflow, compare authored run records under the same rules, and present the evidence behind your claims. Then apply these habits to a project in your chosen major.
Course overview & build checkpoint- 6.1Choose a person and a small problemTurn an interest-pathway idea into a bounded task with a user, permitted inputs, and measurable success.Learn · code · experiment
- 6.2Build the smallest complete loopConnect the input, model or explicit workflow, allowed tools, evidence, and output into a reproducible prototype.Learn · code · experiment
- 6.3Try to break your own resultCompare the prototype and baseline on held-out cases, misleading input, missing evidence, and resource failures.Learn · code · experiment
- 6.4Tell the story of what you builtDemonstrate the agent, justify its architecture, and explain the sources, checks, observed resource use, and limits. Actual API cost needs real usage records.Learn · code · experiment
Your practical outcomeA scoped project brief, a runnable catalog-backed preview, a comparison of authored run records, and an honest project report you can download with your notebook.
Put your skills to work.Choose a project for your major
The courses are your main path. Coding missions and visual explainers give you extra practice when you want it.
8 MAJOR PATHWAYS · GUIDED PROJECT JOURNEYS
What will you
try first?
Choose a field you are curious about.
Understand a problem, learn the tools, then build and test an agent for that field.
Each major has its own two projects. Choose a major to follow its demos, lessons, and builds together.
Major pathway · Two guided projects
Healthcare
Review health-science evidence and keep biomedical claims within what was measured.
Major pathway · Two guided projects
Computer science
Build tools that inspect code and check their own results.
Major pathway · Two guided projects
Law & policy
Compare documents, trace a claim, and explain uncertainty.
Major pathway · Two guided projects
Business & finance
Check a budget and support a campaign's claims with evidence.
Major pathway · Two guided projects
Engineering
Measure a system and test whether a plan fits its constraints.
Major pathway · Two guided projects
Psychology
Ask what a survey can tell you—and what it cannot.
Major pathway · Two guided projects
Arts & media
Prepare a creative project with sources, credits, and review.
Major pathway · Two guided projects
Education
Design useful hints and activities for other learners.
1 / 8 · Healthcare
New to agents? Follow the foundations course. These are broad fields of interest; pre-med and pre-law do not require a particular college major.
SMALL LABS. SERIOUS IDEAS.
Curiosity, meet
real understanding.
This is your place to explore how agents work, what they can do, and where they need limits. Every lesson pairs an original explanation with something you can change and test.
Learn from ideas in research and official engineering guidance—including Google DeepMind, Anthropic, OpenAI, Meta, NVIDIA, Amazon, and the open-source community.
Source organizations are not affiliated with or endorsing AI Lab.
Research behind the curriculumOriginal teaching. Traceable ideas. Clear limits.
Research and official engineering guidance inform the mechanisms. We teach them here through original explanations and classroom exercises. Organization names identify sources, not affiliations or endorsements.
Defending Against Indirect Prompt Injection Attacks With Spotlighting
Research paper · March 2024Separating the provenance of retrieved content and user instructions helps address indirect prompt injection.
How the classroom example differs
The paper evaluates particular mitigations and conditions. A classroom filter or trust flag neither implements the full method nor guarantees protection against all attacks.
Caching challenges and strategies
Engineering guidance · Reviewed 13 September 2026 · undated guidanceReusing observations can reduce dependency calls; cache scope, freshness, and failure behavior need deliberate design.
How the classroom example differs
A room ID is a sufficient classroom cache key only while the task and room facts remain unchanged. Caching does not itself refresh stale information.
JavaScript Guide
Official documentation · Updated 7 November 2025Functions, objects, conditions, loops, and asynchronous code provide the programming foundations for an agent controller.
How the classroom example differs
Ordinary JavaScript rules do not imply that a language model is running. An execution environment may support only part of the language or its host APIs.
Building effective agents
Engineering guidance · 19 December 2024Distinguish fixed workflows from model-directed actions; start simply, use tool observations as feedback, and set stopping conditions.
How the classroom example differs
The article notes that its tooling landscape has changed. Its patterns do not establish that a more autonomous or complex system is better for every task.
ReAct: Synergizing Reasoning and Acting in Language Models
Research paper · 2022 preprint · ICLR 2023Interleave model-generated reasoning and actions with observations from tools or an environment.
How the classroom example differs
A classroom action trace is not private model reasoning. A deterministic teaching loop does not reproduce the paper's models, experiments, or results.
Retrieval-Augmented Generation for Knowledge-Intensive NLP Tasks
Research paper · NeurIPS 2020Combine information retrieval with generation so a model can use external passages while producing an answer.
How the classroom example differs
The paper uses a trained neural retrieval and generation architecture. A keyword lookup illustrates retrieval but neither reproduces that architecture nor proves a retrieved claim true.
Writing effective tools for agents — with agents
Engineering guidance · 11 September 2025Design distinct tools with clear parameters, relevant returned information, and evaluations of how the agent actually uses them.
How the classroom example differs
A description or schema does not guarantee the right action. A live tool can return different data for the same arguments as its environment changes.
The Llama 3 Herd of Models
Research paper · 31 July 2024 · revised 23 November 2024Tool definitions and descriptions guide proposed calls; executed results return to model context. The report covers sequential, nested, and parallel function calls.
How the classroom example differs
This historical model-training report is not a current SDK contract. Generating a call does not execute or authorize it, and benchmark results do not describe classroom performance.
Conversational Context: Session, State, and Memory
Official documentation · Reviewed 13 September 2026 · undated documentationSeparate the current interaction's events and state from searchable information that can span sessions. Choose services according to the required storage lifetime.
How the classroom example differs
In-memory stores lose data on restart. Stored or retrieved information is not automatically true, relevant, persistent, or safe to share between users.
Effective context engineering for AI agents
Engineering guidance · 29 September 2025Curate limited context, retrieve relevant information, and distinguish compaction, external notes, and separate subagent contexts during longer tasks.
How the classroom example differs
Summaries can lose details. Persistent notes do not change model weights or establish truth, and a larger context window does not guarantee perfect recall.
Guardrails and human review
Official documentation · Reviewed 13 September 2026 · undated documentationDistinguish automatic checks from approval decisions, pause sensitive tool requests, retain state, and resume after an application approves or rejects them.
How the classroom example differs
Model-generated approval text is not authorization. Resume examples that automatically approve a request do not establish that a person reviewed it.
Policy in Amazon Bedrock AgentCore: Control Agent Interactions
Official documentation · Reviewed 13 September 2026 · undated documentationEvaluate identity and tool inputs at a gateway before allowing a call. Treat policy authoring, review, enforcement, and decision logging as separate operations.
How the classroom example differs
The gateway governs capabilities routed through it. A classroom approval check is not an AgentCore integration or a complete production authorization system.
Llama Guard 4 Model Card
Official documentation · Reviewed 13 September 2026 · undated model cardClassify model inputs and responses under a content policy, including mixed text-and-image inputs; distinguish classification from other system controls.
How the classroom example differs
The classifier has training-data and knowledge limits and can face adversarial attacks. It does not replace permissions or a sandbox. Model weights have a separate community license.
Demystifying evals for AI agents
Engineering guidance · 9 January 2026Define tasks, trials, and graders; inspect both execution records and final outcomes; repeat trials when model behavior varies.
How the classroom example differs
A score depends on its cases and grading rules. Repeating a deterministic classroom case does not measure the variability of a live model.
Evaluate agent workflows
Official documentation · Reviewed 13 September 2026 · undated documentationUse traces to locate workflow failures, then apply structured graders and repeatable datasets to compare behavior across changes.
How the classroom example differs
Traces record observable execution, not private model reasoning. Passing selected criteria does not demonstrate correctness on every task or connect this lab to an evaluation service.
Agent Evaluation in NVIDIA NeMo Agent Toolkit
Official documentation · Version 1.8 observed · reviewed 13 September 2026Run curated cases, inspect generated answers and intermediate steps, retain effective configurations, and enable profiling when operational measurements are needed.
How the classroom example differs
Different artifacts require different configuration. Scores depend on cases and graders; timing measurements do not establish correctness, and model-based grading is fallible.
Observe your agent applications on Amazon Bedrock AgentCore Observability
Official documentation · Reviewed 13 September 2026 · undated documentationUse instrumented traces and operational measurements to inspect execution, intermediate outputs, duration, errors, and reported model usage.
How the classroom example differs
Monitoring does not enforce a budget or make an answer correct. Tool calls, tokens, elapsed time, and money are different quantities; simulated models provide no real token bill.
Timeouts, retries, and backoff with jitter
Engineering guidance · PDF copyright 2019 · reviewed 13 September 2026Handle transient failures with timeouts and bounded retries, consider backoff and jitter, and establish whether repeating an operation is safe.
How the classroom example differs
Retries can amplify overload or duplicate side effects. Three immediate classroom attempts are an exercise rule, not a universal policy for real services.
LangGraph Graph API quickstart
Official documentation · Reviewed 13 September 2026 · maintained documentationConnect a model node, executable tools, returned messages, and a conditional route that continues the loop or ends the run.
How the classroom example differs
The graph library does not supply a model or guarantee correct calls. The optional local toolkit needs its own model runtime; this browser simulator does not execute LangGraph.
NVIDIA NeMo Agent Toolkit Overview
Official documentation · Version 1.8 observed · reviewed 13 September 2026Compose agent and tool components, inspect workflow measurements, connect MCP tools, and delegate tasks through A2A client/server integrations.
How the classroom example differs
Protocol support does not establish trust or permission. An open-source library does not make every connected model free, local, or available inside a browser.
Tools
Official documentation · Specification revision 25 November 2025Specify tool discovery, input schemas, invocation, and returned results across a client/server connection.
How the classroom example differs
Schemas and annotations do not authorize an action. Clients must not trust annotations from untrusted servers; transport and access controls need separate consideration.
WebArena: A Realistic Web Environment for Building Autonomous Agents
Research paper · 2023 preprint · ICLR 2024Evaluate browser agents in reproducible website environments, checking functional outcomes rather than requiring one exact sequence of actions.
How the classroom example differs
This research benchmark does not represent every website or permission boundary. Classroom examples are not WebArena results, and historical model scores are not current rankings.
SWE-bench: Can Language Models Resolve Real-World GitHub Issues?
Research paper · ICLR 2024 · original benchmarkEvaluate patches against repository issues and executable tests using a reproducible harness. Inspect the actual code change and its tested behavior.
How the classroom example differs
Benchmark variants cover different tasks. Passing a repair case does not establish general coding ability or that a patch meets every unstated requirement.
AlphaEvolve: A Gemini-powered coding agent for designing advanced algorithms
Engineering guidance · 14 May 2025 · research-system articlePropose programs, execute and score candidates, then use evaluation feedback and a program database to guide later proposals.
How the classroom example differs
This research-system description is not a classroom SDK. A bounded practice loop is an adaptation, not a reproduction of AlphaEvolve or its scientific results.
Toolformer: Language Models Can Teach Themselves to Use Tools
Research paper · 9 February 2023 · v1 reviewed 14 September 2026Distinguish model training for selecting API calls and incorporating results from the application code that validates and executes those calls.
How the classroom example differs
Declaring a function does not perform Toolformer's training procedure. Tool-use training does not authorize a call or establish that every proposed argument is valid.
Reflexion: Language Agents with Verbal Reinforcement Learning
Research paper · 20 March 2023 · v4 revised 10 October 2023 · reviewed 14 September 2026Use feedback from an attempted task to form a textual note that can inform a later attempt, without updating model weights.
How the classroom example differs
The feedback or its interpretation can be wrong. A stored note neither guarantees improvement nor fixes a failed service; this lesson does not reproduce the paper's experiments.
Tree of Thoughts: Deliberate Problem Solving with Large Language Models
Research paper · 17 May 2023 · v2 revised 3 December 2023 · reviewed 14 September 2026Generate and evaluate alternative intermediate candidates, then explore them with bounded search and optional backtracking.
How the classroom example differs
Model-generated evaluations are heuristics and can reject a useful branch. Results on the paper's tasks do not guarantee that broader search improves every task or makes a classroom controller a reasoning model.
Generative Agents: Interactive Simulacra of Human Behavior
Research paper · 7 April 2023 · v2 revised 6 August 2023 · reviewed 14 September 2026Connect recorded observations, selected memories, generated reflections, and later plans in a simulated environment.
How the classroom example differs
Believable behavior by virtual characters does not validate claims about real students. A model-generated reflection is an inference, and the lesson's temporary store is not the research architecture.
AgentBench: Evaluating LLMs as Agents
Research paper · 7 August 2023 · v3 revised 4 October 2025 · reviewed 14 September 2026Evaluate agent behavior in distinct interactive environments and keep the task, environment, and evaluation procedure visible when interpreting a result.
How the classroom example differs
Its eight environments do not represent every deployment. Historical model results are not current rankings, and a classroom pass rate cannot be compared with AgentBench results.
SIMA 2: An Agent that Plays, Reasons, and Learns With You in Virtual 3D Worlds
Engineering guidance · 13 November 2025 · research-system article reviewed 14 September 2026Show that agents can act in virtual 3D environments and still require observations, task checks, and explicit limits.
How the classroom example differs
The article identifies challenges in long tasks, memory, visual understanding, and precise actions. This lab does not run SIMA 2, reproduce its training, or establish real-world physical capability.
OpenAI Agents SDK for Python: introduction
Official documentation · Reviewed 14 September 2026 · maintained documentationIdentify the SDK's model-and-tool agent abstraction, execution loop, validation hooks, delegation, sessions, and tracing as separate application mechanisms.
How the classroom example differs
An SDK is not a model or an assurance of correct behavior. This browser JavaScript environment does not install or execute the Python SDK, and a real provider setup has its own requirements.
Agent orchestration — OpenAI Agents SDK
Official documentation · Reviewed 14 September 2026 · maintained documentationDistinguish code-directed control from model-selected steps, a manager calling a specialist as a tool, and a handoff that changes the active agent.
How the classroom example differs
These patterns have different control flow and tradeoffs. Naming several roles does not establish independence, permission, or better results; the classroom routes are ordinary JavaScript.
Foundation: Introduction to LangGraph — Python
Official documentation · Reviewed 14 September 2026 · undated public syllabusIdentify state, reducers, human feedback, parallel work, and subgraphs as distinct concepts to learn when composing an agent workflow.
How the classroom example differs
Only the public syllabus was inspected; enrolled lectures and notebooks were not accessed. Our JavaScript lessons, order, code, and student exercises are original rather than copies of this course.
Graph API overview
Official documentation · Reviewed 14 September 2026 · maintained documentationRepresent work as nodes and edges over state, and choose how each state key combines updates through a reducer.
How the classroom example differs
An update may replace a value rather than append to it unless the configured reducer says otherwise. Drawing a graph does not establish concurrency, correct state merging, or a model connection.
Persistence
Official documentation · Reviewed 14 September 2026 · maintained documentationDistinguish thread-scoped checkpoints from a store used across threads, and select storage according to the desired lifetime.
How the classroom example differs
An in-memory checkpointer is not durable across a process restart. The classroom fixture, the browser's saved draft, and a real framework checkpoint are different stores.
Querying
Official documentation · Reviewed 14 September 2026 · maintained documentationSeparate retrieval, postprocessing of retrieved candidates, and response synthesis rather than treating a search result as a completed answer.
How the classroom example differs
Retrieval or filtering can leave no useful source. A query engine does not establish that an answer is factually supported, and this browser lab does not run the Python framework.
Introduction to Agentic Frameworks
Official documentation · Reviewed 14 September 2026 · undated public course pageCompare the roles of frameworks such as smolagents, LlamaIndex, and LangGraph after understanding the agent loop; simple tasks may use ordinary code.
How the classroom example differs
Framework examples are not interchangeable APIs or evidence that a framework is required. No course exercise, model-performance claim, or certification is reproduced here.
AI Agents in LangGraph
Official documentation · Reviewed 14 September 2026 · undated public indexed course overviewUse a build-first, framework-second learning sequence to connect basic agent mechanics with graph-based orchestration.
How the classroom example differs
Direct course-page access failed during review; only the primary-domain indexed public overview was available. No enrolled lectures or notebooks were read, and our classroom exercise does not reproduce the course.
Build an Agent Improvement Loop with Traces, Evals, and Codex
Engineering guidance · 12 May 2026 · reviewed 14 September 2026Turn observed traces and reviewed feedback into reusable evaluation cases, then check a proposed change against recorded conditions.
How the classroom example differs
The live Python/API example was inspected, not executed. Our authored browser cases do not connect to its services, validate a live-model improvement, or establish educational effectiveness.
Optional practice library10 coding missions, 7 visual explainers, and 5 field simulations · Practice a course skill
Common essentials · Extra practice
Practice the shared coding skills
Short, general exercises in tools, evidence, and testing. Apply these skills in the major project you choose above.
Checking this browser’s progress…
Beginner · 3 missionsGive your code a goalRead an input, call a tool, and follow a simple agent loop.
Intermediate · 4 missionsMake the answer worth trustingCheck evidence, ask permission, and test your agent’s answer.
- Give the answer evidenceRetrieve evidence and check that it supports the answer.6 min · 3 trials
- Put permission in the codePut human permission in the controller before an action.5 min · 4 trials
- Catch the almost-right answerFind an almost-right solution with boundary and failure cases.6 min · 4 trials
- Bring your agent togetherBuild an agent with a loop, evidence, and an approval gate.8 min · 6 trials
Advanced · 3 missionsEngineer for the messy casesHandle tool failures, limit calls, and check a proposed plan.
- Recover from a failed toolRetry a temporary failure and stop when recovery runs out.9 min · 5 trials
- Spend tool calls wiselyReuse observations without spending a tool call twice.8 min · 4 trials
- Separate planning from verificationCheck a proposed plan against independent tool evidence.10 min · 6 trials
Prefer to see an idea first?7 visual explainers · Read, change a simulation, and check your understanding
3 min · Visual lesson 1From answering to actingDistinguish a model, a workflow, and an agent by who chooses the next step.3 min · Visual lesson 2Give the agent a useful briefSeparate what the agent should achieve, what it can use, and what it knows for this task.3 min · Visual lesson 3Watch the loop change directionTrace an action, its observation, and the next decision.3 min · Visual lesson 4Make the answer earn your trustMatch a claim to a relevant source and spot what remains unknown.3 min · Visual lesson 5Test the result, then improve itChoose checks that measure the goal and use a failure to revise the design.3 min · Visual lesson 6Decide what the agent may doSeparate information from instructions and put a human check before consequential actions.4 min · Visual lesson 7Design your first agentBuild a complete plan and explain how it can succeed, fail, or stop.
Take the idea into a field you care about5 guided simulations · Coding, research, law, business, and engineering
Healthcare / Pre-med · 10–15 minWhich discovery deserves another test?A higher reading hides a weaker result. Help a research agent compare the evidence.Computer Science · 10–15 minIt says fixed. Can you prove it?Catch a ticket calculator's sneaky boundary bug. Choose a repair and prove it with tests.Law / Pre-law · 10–15 minWhich clause actually supports that answer?A real citation points to an outdated rule. Find the right version and the missing evidence.Business / Marketing / Finance · 10–15 minCan your agent find the best verified offer?The lowest sticker price can hide the wrong deal. Give your buying agent the complete request.Engineering · 10–15 minImprove the score. Keep the requirements.Zero energy sounds perfect—until the class has no computers. Build an evaluator that catches the shortcut.
A few things before you start
What is an AI agent?
An AI agent is a system that uses a model to choose steps toward a goal, often by calling tools and using their results to decide what to do next. A fixed sequence of programmed steps is a workflow. The first lessons help you tell these apart.
Can I start without knowing how to code?
Yes. Start with the model–tool distinction, then learn JavaScript values, objects, functions, and conditions. Each lesson explains a small idea before you edit a working starter and compare test cases.
Is AI Lab free, and do I need an account?
The lessons and browser coding labs are free to explore without an account or an API key. Curriculum drafts and progress are saved in your current browser. The live workshops are separate programs.
Does the coding lab use a live AI model?
Your JavaScript executes in an isolated runtime. Model choices, school records, and tool responses are authored classroom simulations. A separate optional local SDK project shows real framework integration; it needs its own setup.
What will I be able to build?
You can build and test a bounded tool loop, retrieve evidence, handle errors, and check proposed actions. Then apply those skills to a guided student-life or career-interest project, such as a study planner, evidence helper, or patch reviewer.
Which learning path should I follow?
Start with course 1 if you are new to coding. The six short courses move through foundations, tools, evidence, reliability, agent systems, and a final build. Each course lists its prerequisites, four lessons, and a practical checkpoint. Choose a major project when you want to apply these skills to a field; extra coding missions and visual explainers are optional practice.
Your JavaScript really runs. The model decisions and school data are authored simulations, so you can learn without an API key. Every workspace also includes a separate real SDK example to explore next. Passing the lab’s cases is practice, not proof that an agent is ready for real-world use.