Someone gets handed training responsibilities at work. Maybe they're an HR generalist, a team lead, an operations manager. Maybe their title doesn't have "learning" in it at all. A problem shows up, leadership says "we need training on this," and suddenly they're the training person. No instructional design background, no diagnostic framework, no time to get one. Just a deadline and a vague sense that something is wrong.
I call them accidental L&D professionals, and there are a lot of them.
Here's what I watched happen, over and over. They'd reach for whatever tool was nearest. Increasingly, that meant an AI tool that promised to generate training content fast. They'd type in the request, get a polished course outline back in seconds, build the thing, deliver it, and then watch it not work. The behavior wouldn't change. The metric wouldn't move. And nobody could quite say why.
The reason is almost always the same: the request was wrong. Not the AI's output, the request itself. Someone upstream decided training was the answer before anyone checked whether training could actually solve the problem. And the tools made it easier than ever to skip that check.
This is the part that bothered me. The L&D field has known for decades that most performance problems aren't training problems. They're broken processes, unclear expectations, missing consequences, conflicting incentives, tools that don't work. Training is appropriate only when the gap is genuinely a knowledge or skill gap. Everything else needs a different intervention. But the new generation of AI tools assumes the request is the problem and gets to work generating content. They're making bad training faster, not better training.
So I started building the opposite.
Thinking Forge is an AI-powered thinking partner for training design, built around a diagnostic framework I call the Forge Method. It refuses to generate content until it has helped the user figure out what's actually broken. The first stage is root cause diagnosis. The second is defining what the learner needs to experience. The third is designing the conditions for the learner's own understanding to emerge. The fourth is measurement, with a built-in failure loop that traces flat results back to the specific earlier decision that caused them.
The core philosophy is simple: the learner's brain does the work, and the designer builds the conditions. The tool exists to protect that principle from the pressure to skip ahead.
Thinking Forge sits inside a three-tier model. The L&D Prompt Pack is for people who want better questions. The app itself is for people who want a better process. And Forge Review is a human expert layer for people who want a second set of eyes on a real project. Each tier meets the user at a different point in the same problem.
The app is in active build, with a waitlist open at thinkingforge.app. The Training Audit is live and taking projects. The Prompt Pack is available. I'm posting about the build process on LinkedIn as I go.
The quiet outcome. Not "this tool generated a course in 30 seconds." That's the wrong win. The right win is someone opening the tool, working through the diagnosis, and closing it because they realized they didn't need a training course at all. They needed to fix a process, or have a hard conversation with a manager, or change an incentive. If Thinking Forge talks people out of training when training isn't the answer, it's working.
That's the version of L&D AI I want to exist. So I'm building it.