ANDAMAS LAB / 25 PRACTICAL PATHS
Choose the work you want to improve. Every Path starts with free Foundations; Advanced is also free during Early Access, with an optional Final Lab included when the Path has one.
EXPLORE ALL PATHS
16 of 25 Paths shown.
COLLECTION / WORK
Organize one real project so an AI can find the right truth without being fed every file every time. Build a small knowledge system with clear source hierarchy, durable context and an update routine.
Use AI to investigate a real question without turning search results into a confident collage. Build a research process that separates facts, source quality, uncertainty and the actual decision you need to make.
Create a practical place for AI-assisted work that remembers project context without mixing every client, file and decision together. Design projects, routines, reusable instructions and clear boundaries for what stays human.
COLLECTION / CREATE
Turn a pile of ideas, drafts and channels into an editorial workflow you can actually operate. Build a clear path from source material to brief, draft, review, approval, reuse and measurement without losing what was approved or why.
COLLECTION / OPERATE
Build an AI-assisted operating loop for the sales work that actually moves opportunities: account research, meeting preparation, follow-up, opportunity judgment and weekly prioritization. Use AI to improve commercial judgment and consistency across the week.
Turn partnerships from a folder of interesting companies and sporadic introductions into a repeatable operating process. Research fit, qualify the real exchange of value, prepare outreach, track commitments and review the portfolio without automating relationship judgment away.
Build a competitor and market signal system that tells you what changed and why it may matter — without turning every product announcement into strategic panic. Use traceable sources, comparison frames, change logs and decision triggers.
Design AI-assisted support that helps with repetitive questions while making uncertainty, escalation and human ownership obvious. Build from real request categories, reliable knowledge and safe handoffs instead of dropping a chatbot onto incomplete documentation.
Find repetitive operational work worth automating and redesign it before connecting tools. Build automations with explicit triggers, inputs, human authority, exception paths and evidence that the task actually completed.
COLLECTION / BUILD
Go from a fuzzy website idea to a real deployable site. Define the job of the site, structure the content, build with an AI-assisted workflow, review the result in a browser and publish something you can maintain.
Keep the speed of AI-assisted coding while adding the boring disciplines that stop a prototype from becoming a fragile production mystery. Learn repository truth, architecture boundaries, migrations, secrets, reviewable changes, CI, browser verification and controlled deployment.
Use AI to expand testing capacity without confusing generated tests with evidence that the product works. Build risk-based acceptance, deterministic checks, real-browser verification, regression cases and a release gate that connects code quality to the user story.
Put a small AI-enabled application on your own VPS without turning yourself into a full-time infrastructure engineer. Build a minimal Dockerized stack with private database, reverse proxy, secrets, health checks, backups, secure access and a promotion runbook you can actually operate.
COLLECTION / ORCHESTRATE
Build an MCP server that gives an AI client useful, bounded access to real tools and information without handing the model unrestricted application authority. Start from one job, expose the minimum capability, secure the connection and test the protocol as a product boundary.
Move from one overloaded assistant to a small team of specialized agents only where specialization changes the work. Define jobs, context, tools, handoffs, ownership and shared artifacts so collaboration reduces complexity instead of multiplying prompts.
Build the control layer around agents: capability trust, tool permissions, authority/effect classification, human approvals, compatibility, cost admission, lineage, stop conditions and evidence about what actually ran. Governance should make useful agents safer to operate, not bury them in paperwork.