After spending time in the product trenches building personalised LMS experiences, I've come to believe this architectural choice matters far more than which AI model sits underneath it.

Here's my honest take.

Rule-based pathways: safe, until they aren't

Rule-based pathways are the comfort food of L&D. Predictable, auditable, easy to explain to a compliance team. But the moment your org restructures or a skill set shifts, they start cracking. They were built for a world where roles were stable. That world is gone.

AI-first pathways: exciting and risky

AI-first pathways are genuinely exciting — and genuinely risky. When they work, they feel almost magical: the right content, at the right depth, at the right moment. When they don't, nobody can explain why a learner ended up where they did. In enterprise environments, "I don't know, the AI decided" is not an acceptable answer.

Hybrid is where I've landed — and not as a compromise

Rules set the floor. AI raises the ceiling.

Compliance requirements, prerequisites, mandatory sequences — those stay locked. Within that structure, AI earns its keep by personalising depth, pacing, and focus in ways no human-authored rule tree ever could. Learners get something that feels personal. Admins retain control. And when someone asks "why did the system recommend this?", there's an actual answer.

My honest summary

  • Rule-based = safe, but you'll hit a ceiling fast
  • AI-first = powerful, but enterprise trust is hard to earn back once you lose it
  • Hybrid = harder to build, but the only one that scales with integrity

If you're serving multiple clients, fast-moving skill landscapes, and outcome-driven learning programmes — hybrid isn't a middle ground. It's the strategy.

The real question this space needs to sit with

How do we define the boundary between AI autonomy and human control in learning systems? That line isn't drawn once — it needs to be revisited as trust is earned, data matures, and learner expectations evolve. Getting that boundary right is the actual design challenge. Everything else is implementation.

Common mistakes I see teams make

  • Treating AI personalisation as a feature to switch on, rather than an architectural decision that touches governance, reporting and risk.
  • Letting an AI model reorder mandatory compliance content — regulators do not care that the algorithm thought it was helpful.
  • Failing to log why a learner was routed a particular way, which makes audits and disputes impossible.
  • Assuming your existing rule tree is a good starting point for AI, when it usually encodes an organisation that no longer exists.