Most career development programs are theatre. Here's what replaces them.

I ran engineering teams for years before MentorStack, and the same thing happened every quarter. A strong performer would tell me where they wanted to be in two years. We'd build a plan. Three months later we'd have basically the same conversation, and whatever we'd written down last time was either forgotten or quietly irrelevant.

For a while I assumed I was bad at this. Then I asked around and realised it wasn't a me problem. Almost every company I talked to runs career development the same way: an annual review, maybe a quarterly check-in, a development form that lives somewhere on a SharePoint nobody opens. The form-filling becomes the work. Whether anyone actually grew got separated from whether the program ran.

I think that whole model is about to fall apart. And I don't think most HR leaders see it coming.

What changed

You can now open Claude or ChatGPT, type "build me a 12-month plan to move from senior IC to engineering manager," and get back something genuinely good in 30 seconds. Better than what 90% of managers will produce in a development meeting. Not perfect, not personalised to your company, but better.

I noticed this sometime in 2024. I was messing around with one of the early models and asked it to coach me through a career question I'd been chewing on for months. The answer was better than my last performance review. That bothered me at the time. It's the thing that eventually clarified what I'd been getting wrong about the development problem for a decade.

The hard part of helping someone grow has never been writing the plan.

The actual problem

Marketing person on your team wants to be a director in twelve months. The plan part is straightforward. Own a cross-functional campaign, build management foundations, lead a small team, position for the conversation. Most senior marketers can sketch that on a napkin. AI does it in seconds.

What's hard is everything from there. Who should this person learn from inside the company. Who's actually made the same jump in the last two years. Who's got political capital to sponsor them when the seat opens up. Who's running the kind of project they should shadow next quarter.

This is where ambition dies. Not in the planning. In the awkward DM the employee never sends. The introduction the manager promises in a 1:1 and then forgets. The mentor relationship that gets suggested and never started.

Nobody in the development industry wants to admit this, but the planning was never the bottleneck. The execution was. We built a whole industry of frameworks and templates and quarterly reviews around fixing the wrong half of the problem.

The MCP thing, briefly

Quick aside for readers who aren't deep in the AI tooling stuff. MCP (Model Context Protocol) is a standard Anthropic released in late 2024 that lets AI agents call into external systems through a common interface. OpenAI, Google, basically every major AI vendor adopted it within a year. Think of it as the wiring between an agent's reasoning and the tools where actual work happens.

I'd been watching MCP since it was announced and the obvious move kept staring at me. Agents can produce career plans on demand. Mentorship platforms hold the data those plans need to become real: skills, aspirations, the network of who's actually done what inside a company. Connect the two and the gap collapses.

So we built it. The MentorStack MCP server lets any AI agent your employees already use call directly into your mentorship program. Employee asks for a plan. Agent produces one. Plan gets handed to MentorStack. We pull what your org already knows about that person, find the right mentors and sponsors and peers, book the first conversation. Twelve seconds.

What this isn't: an AI coach. There are dozens of those, most of them are bad, and the coach-replacement angle misses the point entirely. AI is good at planning. Humans are good at growing other humans. Let each do what they're good at. The MCP server is just the wiring that makes that division of labour work.

Why this scares HR leaders

When I show this to People leaders, the technically curious ones get it immediately. The hesitation is from somewhere else.

If career planning becomes something an employee does with their AI agent on a Tuesday afternoon, a bunch of the development infrastructure HR has built becomes irrelevant. The annual review cycle gets less central. The development form is just dead. The L&D team's job as the curator of growth pathways changes shape.

I get why that's uncomfortable. I'd push back on the conclusion most HR leaders draw though.

The job doesn't get smaller. It gets sharper. The strategic work shifts from "do we have the right development frameworks" to something more like "is our mentor network deep enough to absorb continuous demand instead of quarterly demand, and do we have the right humans available when employees actually need them." That's a more interesting problem, not a smaller one. But it does mean letting go of work that used to feel central to the function. That's hard. People have built careers around the work that's about to stop mattering.

Some HR leaders will see this and act. Most won't. The ones who act will be ahead in a way that's hard to catch up to, because development is one of those domains where small leads turn into large ones quickly. Your top performers stop leaving because they can develop themselves on demand against the actual humans inside your company. The mentor network grows because people stick around long enough to mentor the next cohort. Eighteen months in, you're meaningfully ahead.

What I'm claiming, plainly

The big one: AI-generated career plans are already better than what most companies produce internally. Not because AI is magic. Because internal career planning is usually rushed, generic, and constrained by the planner's own narrow experience. An agent has read everyone's experience. That's a real advantage and pretending it isn't is the kind of thing companies say right before they get disrupted.

The corollary: the bottleneck in talent development was never planning quality. It was the friction between plan and human network. Twenty years of frameworks and templates have tried to fix planning. None touched the friction.

Which leads to where I'm obviously biased. Connecting AI agents to mentorship platforms through MCP is how the friction gets removed. The architectural argument doesn't actually depend on whether MentorStack wins. It depends on whether AI-produced plans plus human execution is the right pattern, and whether MCP is the right way to wire them together. Yes and yes, in my view.

So

If you run People or L&D, the question isn't whether AI will reshape career development. It's already happening. The question is whether you want to be the team that builds the infrastructure for it inside your company, or the team that watches it happen.

If you want to see what it looks like in practice, the MentorStack MCP server is live: mentorstack.co. We'll walk you through it.

The future of work conversation has been mostly about which jobs survive AI. I think the more interesting question, and the one that determines which companies win the next decade, is what AI does to growth itself. I've got a strong opinion. I'd actually like to hear yours.