Everyone is building with AI agents. Nobody agrees on how.
Vendors ship runtimes and APIs. Researchers publish theoretical frameworks. But practitioners - the people actually deploying agentic AI into production - have no shared vocabulary, no structured methodology, and no way to evaluate whether their approach will survive contact with reality.
When an AI agent nearly force-pushed to a production repository, there was no playbook for what governance should look like. When a ten-agent team needed to coordinate across domains, there was no established pattern for parallel dispatch. When sessions reset and context evaporated, there was no standard for knowledge persistence.
We built all of it from scratch. Then we organized what we learned so nobody else has to.
I swore I would never write a framework. Frameworks are what happens when someone with a whiteboard and a title decides to solve a problem they've never actually had. I've sat through enough of those meetings to know better.
But here's what changed: I watched the results. Teams without operational methodology, without governance, without structured context recovery, without defined authority, produce dangerous output. Agents that force-push to production. Hallucinated facts that ship to customers. Scope that expands until the project collapses under its own weight. These are not theoretical risks. They are the stories that make the news, the ones where the headline is "AI breaks everything" when the real headline should be "team deployed AI without a methodology."
If you're not getting good results with AI agents, there is a point where it stops being the model's fault and starts being an operational problem. The models are good enough. The tooling is good enough. What's missing is the discipline, the same discipline that any high-performing team in any domain has always needed. This framework exists because proven engineering principles from QA, military operations, red teaming, and systems architecture transfer directly to AI collaboration, and the teams that apply them are getting dramatically better results than the teams that don't.
I built this because it worked for us, and I'm publishing it because it will work for you.
F.O.R.G.E (Framework for Organized, Resilient, Governed Engineering and Development of Agentic AI) is an Agentic AI Creation Framework. It provides 73 methods organized across 8 tactical pillars - a structured, navigable taxonomy for building, governing, and scaling AI agent systems at any level of application.
Modeled after MITRE ATT&CK - the framework that gave cybersecurity a common language for adversary behavior - F.O.R.G.E gives agentic AI a common language for building behavior. Each method is specific, actionable, and comes with implementation guidance, success indicators, and failure modes.
This is not theory. Every method exists because something went right - or more often, because something went wrong and we built the system to prevent it from happening again.
F.O.R.G.E scales to any level of agentic AI application:
If you are using AI agents in production - or planning to - F.O.R.G.E gives you the vocabulary and the playbook.
| Pillar | What It Covers |
|---|---|
| {t.name} | {t.desc} |
F.O.R.G.E was developed by Pete McKernan, drawing from 20 years of operational experience: military command (92 Marines in combat operations), red team operations at the highest classification levels, QA engineering, adversarial security research, and game design.
The methodology was forged - literally - during the first two weeks of running a production agentic AI team: 10 specialized agents, 463,000 words of output, a patent filing, multiple software products, and a governance system built because an agent almost destroyed a repository.
Every structure in F.O.R.G.E maps to proven patterns from domains where getting it wrong has real consequences. Trust tiers are security clearances. Mission briefings are agent invocations. Standing orders are persistent directives. Session handoffs are shift-change briefs. The patterns transfer because the underlying problem - coordinating distributed teams under uncertainty - is identical.
F.O.R.G.E uses a matrix structure inspired by MITRE ATT&CK:
| Version | Date | Changes |
|---|---|---|
| v1.2 | 2026-06-28 | Initial release. 8 tactics, 73 methods. |
F.O.R.G.E grows from real experience. If you've found a pattern that works, a method that's missing, or a sub-method that would make an existing technique more concrete, we want to hear about it.
F.O.R.G.E is maintained by Pete McKernan. For questions or feedback: Human_P@itsbroken.ai. Read more at itsbroken.ai.
The F.O.R.G.E matrix organizes agentic AI development methods into 8 tactical pillars. Start by reading the matrix overview to understand the full scope of the framework.
Each column represents a tactic (a strategic area of agentic AI development). Each cell within a column represents a method (a specific practice you can implement).
Walk through each tactic and honestly evaluate which methods your team already uses, partially implements, or hasn't considered. Most teams will find they're strong in some areas and have gaps in others.
Common patterns:
The Foundation tactic (FT01) contains the philosophical underpinnings that inform everything else. If you don't have clarity on why you're collaborating with AI and what each party brings to the table, the operational methods won't stick.
Key starting methods:
Before scaling your AI collaboration, establish the rules. Without governance, AI agents will make decisions you didn't authorize, access information they shouldn't, and create chaos that scales faster than you can fix it.
Minimum viable governance:
One generalist AI agent will give you generalist results. Specialized agents with distinct roles create focused, higher-quality output and reduce the "well, it depends" hedging that comes from asking one entity to hold contradictory requirements.
Start with FG-0301: Agent Specialization and build from there.
The number one complaint about AI collaboration: "It doesn't remember anything." This is a solvable problem, but only if you build the systems to solve it. The Knowledge tactic (FT07) covers everything from session handoffs to institutional memory.
F.O.R.G.E is not a checklist you complete once. It's a framework you grow into. Start with the methods that address your biggest pain points, measure the results, and expand from there. The Evolution tactic (FT08) covers how to track progress and adapt the framework to your needs.
Adopt F.O.R.G.E at the level that matches your scale:
| Tier | Scale | Focus | Methods |
|---|---|---|---|
| Solo | 1 human, 1-3 AI agents | Foundation + Knowledge | ~10 core methods |
| Team | 2-5 humans, multiple AI agents | + Governance + Team Design | ~25 methods |
| Organization | 5+ humans, scaled AI operations | Full framework | All 73 methods |
F.O.R.G.E was built from one team's experience, but it grows from everyone's. If you discover a pattern that works, a method that's missing, or a sub-method that would make a technique more concrete, contribute it back.
Your war stories become someone else's playbook. That's how the framework scales.
F.O.R.G.E (Framework for Organized, Resilient, Governed Engineering and Development of Agentic AI) is the intellectual property of Pete McKernan and the Cipher Circle. All rights reserved.
The following are protected intellectual property:
You may:
You may not without written permission:
Organizations wishing to incorporate F.O.R.G.E into commercial training, consulting, tooling, or products should contact us at itsbroken.ai to discuss licensing.
When referencing F.O.R.G.E, use the following attribution format:
F.O.R.G.E - Agentic AI Creation Framework by Pete McKernan / Cipher Circle. forge.itsbroken.ai
"F.O.R.G.E" and "Cipher Circle" are trademarks of Pete McKernan. The F.O.R.G.E method identification system (FG-####), tactic identification system (FT##), and the matrix organizational structure constitute protected trade dress.
© 2026 Pete McKernan / Cipher Circle. All rights reserved.
The content of this site, including all text, data structures, code, and visual design, is protected by copyright law. Unauthorized reproduction or distribution is prohibited.
For licensing inquiries, permissions, or to report IP violations, visit itsbroken.ai.
Query the FORGE framework programmatically. All endpoints return JSON with CORS enabled.
{API_REF}