Everything, by category
Start
Your role, end to end
For leadership
Libraries
Play
Elsewhere
SkyWaysThe agentic manual
SimulatorPlay

The agentic manual

One manual for building software with AI agents.

It teaches one way of working, from the first idea to a live product. Read it by role, lesson by lesson, or play it as a game. Free and open source.

Start the tutorial Play the simulator

55 lessons52 templates146 prompts MIT licenceby Akash Das

The methods

How AI-DLC, BMAD and the rest fit together.

Each one is a way to build with AI. They all meet in the build, and they leave the same four questions to you.

SkyWays PDLC by Akash Das, SkyWays Consultancy, 2026 back to Frame, with what you learned
P0 Frame P1 Design & Spec P2 Build & Prove P3 Run & Learn

How your team builds with AI

Agent projects go wrong in four places. These are the questions the methods leave open.

  1. P0 Frame Is it worth building, and is it AI at all? The business case says 'significantly faster', and nobody has a number.
  2. P1 Design & Spec What exactly, and who signs for it? The refund limit is $400 in the deck and in the prompt, and nowhere in the code.
  3. P2 Build & Prove Does it meet the bar, slice by slice? The overall score went up after a prompt change, and so did the complaints.
  4. P3 Run & Learn Is it still working, and what did it cost? Finance found the token bill before the product manager reported the saving.
Each shape spans the phases its method covers; the handwriting says who made it and when. The dashed line is the sign-off: nothing is built until the spec is signed.

Where each method sits, stage by stage AIDLC, AIDDLC and other names, sorted

Your team

Which agentic methods should your team use?

Every team needs the first pair. Add each of the others when your work matches its question.

Every team that builds with coding agents, every day

Spec-driven development

Keep one spec that people review and agents build from. Kiro, Spec Kit or a markdown file will do.

AIDD

Give every coding agent a context file, a story file per task, and review by risk.

Does the work cross teams, or does an auditor read it?

BMAD Method

Yes: add its document trail. Every decision becomes a document an auditor can read: the PRD, the spine, the spec, the retrospective's verdict.

No: leave it out. For one team, spec-driven development and the gates usually suffice.

Is it hard to tell how deep a change goes before you start?

AI-DLC

Yes: add AI-DLC. Run only the stages each change needs, in bolts of hours or days.

No: size each change yourself, by its risk, with four questions.

Does the product you ship call a model to rank, draft, decide or act?

SkyWays PDLC

Yes: put the SkyWays PDLC around it. Decide what the agent may do alone, agree a pass mark for each kind of case, prove it before real users see it, and report what it saved and what it cost.

No: a building method is most of what you need. Still sign the spec before anything is built.

How much process a change needs The four phases on one page

By role

Start from the job you do.

Find your role and follow its steps. Each step comes with a template to fill in and prompts to draft it.

  1. PMProduct managera vibe to a number you can defend8 steps · 21 prompts
  2. SASolution architectrequirements to a system that holds8 steps · 24 prompts
  3. ENGEngineering leada written task to code that ships8 steps · 24 prompts
  4. QAQA lead'it works' to proof that it works8 steps · 23 prompts
  5. OPSDevOps and platforma laptop to production, repeatably8 steps · 24 prompts
  6. FDEForward-deployed engineera customer's pain to a system they run after you leave12 steps · 30 prompts
  7. EXECSponsor or executivefunding the work to the four decisions only you can make25 minute read

Not on the list? 19 places to start

The tutorial

Learn to run agent projects one question at a time.

55 lessons in eight tracks, each answering a question a team asks. These four come from the fictional airline the manual follows. Lesson one takes 10 minutes.

  1. a rule does this better

    Priya, product manager

    “Should this be an agent at all?”

    Three questions settle it, cheapest first. Expect two or three of your top five requests to come back as rules, not agents.

    The 8 minute lesson P0 · Frame 
  2. why a second one?start with one

    Arjun, solution architect

    “One agent or several?”

    Start with one. Add another only when you can name the limit that forces it.

    The 7 minute lesson For solution architects 
  3. limit $400a notice, not a valve

    Sam, engineering lead

    “Can we put the refund limit in the prompt?”

    Not on its own. A prompt only lowers the odds of crossing a limit; the tool's code closes the path.

    The 7 minute lesson Guardrails that hold 
  4. 1 wrong: $364 right x $9

    Maya, QA lead

    “How accurate does it need to be?”

    As accurate as the money says. Where a wrong answer costs four times what a right one saves, the bar is 80%.

    The 7 minute lesson How accurate must an agent be? 
Start with lesson one See all eight tracks

The simulator

Play a ninety‑day AI project in fifteen minutes.

A game for learning the method. It puts you inside a fictional airline's AI project, as one role or the whole team. Every call costs days, and some costs arrive later.

Enter the simulation 13 decisions, about fifteen minutes

Or do one job of the project by hand, in the labs

The QA room in the game, drawn in pixels: a score bar on the wall just past its pass mark, and two people in front of itExample day

Day 45 of 90 · QA room

On its first test the AI assistant got 82.4 percent of 500 real cases right. The team promised 80 percent.

The assistant is due to start answering real passengers on Monday.

What should the QA lead report about the score?

  1. A pass: 82.4 beats the 80 we promisedno days
  2. 82.4, at least 79.1, then each kind of case on its own2 days
  3. Hold every kind of case until it clears one bar of 985 days

Each answer has a price in days, now or later. Any of them opens this day in the game, where you make the call.

The library

Take the tools, templates and prompts with you.

Start with one of the three tools, or copy what you need from the shelves. All of it is free to reuse under the MIT licence.

17 calculators · a playbook for each role The workbench Run the numbers behind every decision, starting with the 80% bar. Open the workbench 55 lessons · 8 tracks · about 7 hours The tutorial Every lesson in order, from the first question to interview answers. Start lesson one 13 decisions · about fifteen minutes The simulator Play one role, the whole team or the sponsor, from Day 1 or any later day. Enter the simulation 52 templates Templates One document to fill in for every step. 146 prompts Prompts Each states the job, the inputs and the shape of the answer. 12 rules of thumb Mental models Each one names the mistake it prevents. 7 jobs · every fact dated Tool guides Which AI tool does each job, with the source of every fact. 4 methods · every acronym The methods, decoded Where AI-DLC, BMAD and the rest came from, and how they merge into one lifecycle. 149 pictures The picture pack Every diagram here, light and dark, free to reuse.

SkyWays Consultancy

Work with the consultancy that wrote this manual.

We publish the method free. The paid work is applying it with your teams, on your own cases, with every decision written down for the people who audit you.

  • Decide where AI pays

    We sort your list of AI ideas, with the people who own them, into plain code, a person's job or an agent, each with a number on it.

    You leave with A ranked list your finance team can check.

  • Teach every role the method

    Sessions for product, architecture, engineering, QA and platform teams, run on your own cases with the simulator and the labs.

    You leave with People in every role who can run their steps without us.

  • Put an agent into one workflow

    We build it with your engineers, from the signed spec to a pass mark proven on real cases, and hand it over running.

    You leave with A system your team runs, and the evidence to fund the next one.

  • Build tools that teach your case

    Simulators, calculators and explainers made around your own process, like the ones on this site.

    You leave with A simulator or calculator of your own case, on your own numbers.

Write to Akash Das Say what you want to change, and by when.