Design training that sticks, audit and improve existing courses, and back every decision with cited research. Learning Brain plugs into any AI tool and equips it with a world-class learning design team.
Ask ChatGPT or Claude to design a course, and you'll get tidy objectives, stock activities, a "knowledge check" at the end. It'll look polished and professional, but if you dig deeper you'll find it falls apart under scrutiny and doesn't pass the "would this work in the real world?" filter.
You can tell your AI to use learning science, and feed it research. But then it'll over-focus on what you've given it, and ignore other things that are just as important. You'll get the same shallow design with better vocabulary. It doesn't know what good looks like. That's the gap Learning Brain closes.
Connect Learning Brain to the AI you already use: Claude, Codex, ChatGPT, or any MCP-compatible tool. It'll take two minutes.
From then on, your AI can call on a team of five learning-science specialists whenever you ask it for learning design. Your workflow doesn't change, but what comes back is sharper: designs grounded in research, pushback when something won't work, and a citation for every decision.
Evaluates whether a learning-science claim is supported by research, and how strongly. Explains the evidence, not just the conclusion. Diagnoses why a running course isn't producing results. Settles design arguments like explicit instruction versus discovery learning for a specific audience.
Designs full course architectures with cognitive load budgets and retrieval schedules. Builds modules against Merrill's five principles. Sequences content so concepts build on each other instead of piling up. Designs adaptive learning paths that route learners based on demonstrated capability.
Writes learning objectives you can assess. Builds MCQ items where every distractor targets a specific misconception. Creates worked examples with fading sequences. Writes feedback that corrects the thinking, not the answer.
Designs live workshop run sheets with real facilitation moves. Specific questioning sequences, not "discuss in groups." Timed segments. Recovery moves for when the discussion dies. Scripts facilitator guides matched to the facilitator's experience. Diagnoses why a specific learner is stuck.
Reviews modules against 12 structural criteria. Audits objectives and MCQs for construct validity. Scans for the ten instructional illusions. Predicts whether a design will transfer to the workplace, and recommends specific bridging moves if it won't.
Plus five cross-cutting tools for eliciting learner context, structuring course briefs, pushing back on badly-framed requests, showing relevant case studies, and tracing citation trails.
I asked four AI models to design a self-paced course on workplace sustainability, giving each the same brief.
Claude Opus (the most powerful model) produced a beautiful interactive HTML course in one go. It looked great, but there was no retrieval practice, no assessment, and no transfer design. Claude Haiku (least powerful model) produced a generic outline with vague objectives. Competent templates that teach nothing.
Every model now included cumulative retrieval practice, cognitive load budgets, worked-examples, transfer plans, post-training retrieval schedules. Amazingly, the smallest, cheapest model produced the most comprehensive output, including success metrics, implementation plans, and manager talking points. This suggests that giving the AI the right information can matter more than the raw reasoning power of the AI itself.
Brief to build-ready pack in one sessionObjectives, architecture, retrieval schedule, assessment blueprint — in one flow.
Catch structural problems before productionAudit a module and get specific, evidence-cited fixes.
Make junior designers saferConsistent quality standards that don't depend on who built it.
Defend design decisions to stakeholdersCited research findings instead of "best practice."
Convert formats without losing the designLive workshop to self-paced, or vice versa.
Review assessments for construct validityQuiz items, question banks, diagnostics that actually measure capability.
Build reinforcement that sticksSpaced retrieval plans, manager talking points, implementation intentions.
Turn an SME deck into real learningStructured experience with objectives, practice, and assessment.
Your team produces dozens of courses a year. You can't personally review every one at learning-science depth. Learning Brain gives you a consistent quality standard across the team. The output stops depending on which designer built it, how experienced they are, or how much time they had. Junior designers get the same research-based checks as senior ones. Senior designers move faster with evidence at hand. Every design decision has a citation trail you can show anyone who asks.
When a VP asks "why is the course structured this way?" you need more than "best practice." You need a specific research finding. Learning Brain gives you the evidence trail. And if your design has a structural problem, it tells you before the VP does.
You're selling expertise, not just production. Learning Brain lets you quality-check deliverables at scale, back every recommendation with cited evidence, and catch instructional illusions before they reach the client. It's the difference between a polished deck and a defensible one.
Your audience trusts you because of what you know. Your course has to actually deliver it. Polished production and five-star reviews aren't the same thing as students getting real results, and the creators who build durable audiences are the ones whose students do. Learning Brain gives you what a dedicated learning designer would bring: objectives that force you to be specific about capability, assessments that tell you whether it's landing, and design that makes your expertise transfer outside the course.
The output looks good but feels hollow. Beautiful outlines, professional language, zero retrieval practice, no transfer design. You sense something's missing but you can't name it. Learning Brain names it, and fixes it. Every design decision comes with the research behind it. The result is better courses now, and better instincts over time.
Every design decision is traceable to a specific research finding. The knowledge base covers nine domains of learning science, and every finding carries an evidence-strength rating — moderate evidence is never laundered as strong. When the research is silent on a topic, the tools say so rather than fall back to general AI knowledge.
AI models are trained to be agreeable. They soften criticism, pass work that should fail, and tell you what you want to hear. Learning Brain pushes back on that in three ways.
Every citation comes from the knowledge base. No fabricated references. Popular myths (learning styles, left/right brain, the 10,000-hour rule, the Cone of Learning) get flagged the moment they appear.
Promises that training will transfer to wildly different situations. "Everyone at the company" as an audience. A 90-minute module covering 35 topics. A cramming schedule. An auto-graded quiz for a skill only a human can judge. Try asking for something like this, and Learning Brain explains why it won't build it.
Nineteen of the 32 tools give your AI a quality rubric and tell it to score its draft, fix what falls short, and only then hand it back. That's strong guidance rather than a guarantee, because how closely it's followed depends on your AI. The refusals above are the checks it can't skip.
I fed the tool 25 inputs designed to fool it, and ran each one on two AI models: modules that look polished but have no demonstration phase, objectives that sound active but promise nothing, MCQs where the longest answer is always correct, engagement-first workshop briefs. Every flaw was caught, and every fix was specific and actionable. New patterns that slip through get added to the refusal list.
The evidence base was built by a practitioner with twenty-plus years of work across educational publishing, corporate L&D, and edtech. Every tool exists because it solved a real design problem, and every rubric exists because a specific failure mode kept showing up in practice. This is the difference between a research database and a working knowledge base.
Mostly no. In Claude you add a connector URL in settings and upload a small skill file. ChatGPT users set it up as a plugin in Codex mode, in the ChatGPT desktop app. No coding, and you don't need an API key for your AI. It takes a couple of minutes, including a confirmation email the first time you connect. The setup guide has one path per tool.
Learning Brain is an evidence layer that plugs into Claude, ChatGPT, or any MCP-compatible AI tool. It gives your AI access to a curated knowledge base of learning science research and 14 quality rubrics. The output is design artifacts: objectives, course architectures, module audits, assessment items, facilitation guides. It doesn't produce finished slides, videos, or SCORM packages. It checks the design against the research before anything gets built, so production work is less likely to be wasted.
It works with the AI you already use. Learning Brain is an evidence layer. It feeds your AI curated research and quality rubrics. Your AI still produces the output. Learning Brain gives it the evidence and the quality criteria to produce it well.
Your AI does the writing. Learning Brain supplies the expertise, and the writing runs on your existing Claude or ChatGPT subscription, so you don't need a separate AI API key.
It produces design artifacts: objectives, course architectures, module structures, assessment items, facilitation guides, audit reports. It doesn't produce finished slides, videos, or SCORM packages. Think of it as the architect, not the builder. Your production team still builds the thing.
Any tool that supports the Model Context Protocol (MCP): Claude (desktop and Code), Codex, ChatGPT, Cursor, Windsurf, and others.
Claude gives the best experience. It picks the right tool for the job, chains tools together, follows the inlined rubrics, and cites sources by name.
For ChatGPT users, Codex mode in the ChatGPT desktop app is the recommended path, where your plan allows plugins. It's the closest to the Claude experience: it chains tools and draws on the research, though it sometimes skips the audit steps.
ChatGPT chat works but is lighter. Custom connectors there are a Business, Enterprise and Edu workspace feature. It tends to call one tool and then summarise, rather than chaining tools together, so it's best for one-off questions. For full audit and design workflows I recommend Claude or Codex mode.
This is common in enterprise orgs. Three paths usually still work even when chat-app connectors are blocked:
1. Claude Code / Cowork — terminal CLI or the Code tab inside Claude Desktop. Separate policy surface from Claude chat, and plugin install is usually allowed. See the Claude Code setup.
2. Codex mode in the ChatGPT desktop app — if your plan allows plugins. Plugin settings are separate from ChatGPT chat's connector settings, so this sometimes works when connectors are blocked. See the Codex mode setup.
3. Claude Desktop via local config — edits a local file on your Mac and bypasses the "Custom connectors" setting entirely. See the restricted-org fallback at the bottom of the setup guide.
If none of those work, email me — I'd like to hear about it.
The research covers how people learn — not what they learn. The principles of cognitive load, retrieval practice, transfer design, and assessment alignment apply whether you're designing compliance training, technical certification, leadership development, or onboarding. The tools adapt to your audience and context.
Good! That's a conversation worth having. The pushback is always grounded in a specific research finding, which you can evaluate against your contextual knowledge. Sometimes the research finding wins. Sometimes your practitioner judgment wins. Either way, the decision is now explicit rather than assumed.
No. The tools handle the science: objectives, sequencing, assessment alignment, transfer design, cognitive load. Your people handle the context: reading rooms, navigating politics, knowing which research finding matters most for this specific audience at this specific company. No tool can guarantee learning outcomes. What Learning Brain does is check the design against research-based criteria and tell you where it falls short. What happens in delivery is still up to humans.
I built Learning Brain because I wanted it to exist, and I wanted to use it myself. It worked well enough that I turned it into a tool. Your AI does the writing, so Learning Brain is free to use.
Platforms like 360Learning, Docebo, and Absorb are full-stack systems for delivering and managing training. Learning Brain is a different category – it's an evidence layer that sits upstream, plugging into the AI you already use to ensure the design quality of what you create before it reaches any LMS.
When you ask ChatGPT a learning-science question, it draws on general training — which includes outdated findings, debunked myths (like learning styles), and confident claims without evidence ratings. Learning Brain draws on a curated, evidence-rated knowledge base with validated quality rubrics. It also refuses to fall back on general knowledge when its evidence base is silent.
Enter your email and you'll get access immediately. Connect to your AI tool and start designing in two minutes. The first time you connect, we'll email you a link to confirm it's you.