Compare · as of October 5, 2026
The leading ways to build with AI agents. Which one is yours?
Compare AI development methods for Claude Code and Codex, from spec-driven toolkits like GitHub Spec Kit and BMAD to skill libraries like Superpowers. Here is what each one believes, who it suits, and where the Enso Method chose differently.
This guide is written by EnsoDynamics, which makes the Enso Method. We compare our method with others as fairly as we can, and we recommend another method when it fits you better.
Not sure where to start?
Answer twelve quick questions about how you build, and we’ll point you to the method that fits, even when it isn’t ours.
About two minutes · Skip any question · No sign-up
Where each method sits
How much of the work each method defines before the build, and how much it runs on its own. Positions are our reading of each method’s own skills and documentation; close neighbours are nudged apart so each one shows.
- Superpowers: planning toward “One change at a time”; steering toward “It runs on its own”
- agent-skills: planning toward “One change at a time”; steering toward “You steer each step”
- Matt Pocock’s skills: planning toward “One change at a time”; steering toward “You steer each step”
- GSD: planning at the “A body of work defined in full up front” end; steering toward “It runs on its own”
- Compound Engineering: planning at the “One change at a time” end; steering toward “It runs on its own”
- BMAD Method: planning toward “A body of work defined in full up front”; steering in the middle
- AWS AI-DLC: planning toward “A body of work defined in full up front”; steering at the “You steer each step” end
- HumanLayer QRSPI: planning at the “One change at a time” end; steering toward “You steer each step”
- GitHub Spec Kit: planning toward “One change at a time”; steering toward “You steer each step”
- OpenSpec: planning at the “One change at a time” end; steering toward “You steer each step”
- gstack: planning at the “One change at a time” end; steering in the middle
- pstack: planning at the “One change at a time” end; steering at the “It runs on its own” end
- Agentic Coding Flywheel: planning at the “A body of work defined in full up front” end; steering toward “It runs on its own”
- The Enso Method: planning at the “A body of work defined in full up front” end; steering at the “It runs on its own” end
Hands-off builds
You agree a plan, then the agent builds alone, often for hours. The rigor is in the build itself: test-first rules, a fresh reviewer per task, QA in a real browser.
Superpowers
Jesse Vincent (Prime Radiant)
Agree the design, then a strictly test-first build with a fresh reviewer and no stalls.
Read the comparisonCompound Engineering
Every (Kieran Klaassen and Trevin Chow)
Plan, build and review, then write down what the work taught, so the next plan starts smarter.
Read the comparisongstack
Garry Tan
Challenge the idea, review the plan from every seat, test it live, then ship it and watch production.
Read the comparisonpstack
Lauren Tan (Cursor)
No spec but the code. Prototypes settle the questions, agents build in parallel overnight, and another agent verifies every change before it lands.
Read the comparisonHands-on at every step
You make each call and review the work as it goes, step by step.
agent-skills
Addy Osmani
A senior engineer’s lifecycle from spec to launch, with you approving each phase and a rollback plan before anything ships.
Read the comparisonMatt Pocock’s skills
Matt Pocock
Be grilled until every decision is yours, then build in small, test-first slices with skills you can adapt or drop.
Read the comparisonHumanLayer QRSPI
HumanLayer (Dex Horthy)
Align with the agent on the product, the architecture and the code’s shape before it writes, then read every slice it builds.
Read the comparisonSpec-driven
The spec is the main deliverable, written to a set format, such as numbered requirements or WHEN/THEN scenarios, and reviewed before any code. The rigor is in the spec; the build follows it.
BMAD Method
BMad Code, LLC (Brian Madison)
Plan only as much as the work needs, keep one living spec at the hub, and build one story per session from a written plan.
Read the comparisonGitHub Spec Kit
GitHub
Spec-first, one feature at a time, under a project constitution, on almost any coding agent.
Read the comparisonOpenSpec
Fission AI (lead maintainer Tabish Bidiwale)
One living spec of what the system does now; each change is a small diff to it, merged in when the work is done.
Read the comparisonWhole-project planning
Plan the work in full first, as a roadmap or a set of requirements, then build it stage by stage.
GSD
open-gsd, the community continuation of TÂCHES’ original
Plan the whole product as a roadmap, then build it phase by phase with fresh-context agents, from step-by-step to fully autonomous.
Read the comparisonAWS AI-DLC
AWS Labs (Amazon Web Services)
A governed lifecycle from idea to operations; specialist agents draft each stage, and a person approves it before the next begins.
Read the comparisonAgentic Coding Flywheel
Jeffrey Emanuel
Plan the whole system with several frontier models, turn it into hundreds of self-contained tasks, and let a swarm of agents build, review and test it.
Read the comparisonThe Enso Method
EnsoDynamics · ours
Define the work in full before the build, and keep it true as the system changes. Open questions become proposals the agent builds on, so the work keeps moving whether the person who decides is at the keyboard or not.
How the Enso Method worksRead a few and still torn? Twelve questions will narrow it down.
Which method fits you?