Just-in-Time Teams: The End of Permanent Headcount

Hey now.

The Idea

Manufacturing figured this out decades ago. Toyota didn’t stockpile parts in warehouses — they built a system where components arrived exactly when they were needed on the assembly line. No excess inventory. No idle capital. Just-in-time.

Software companies never got the memo. They hire 50 engineers and hope 50 engineers’ worth of work materializes. When it doesn’t, you get what every large tech company is discovering right now: structural overhead that AI is about to make indefensible.

The same principle applies to teams now. Just-in-time teams.

How It Works

The core of your operation is AI. Agents that write code, run tests, manage infrastructure, handle deployments, monitor systems. They don’t take PTO. They don’t need to be convinced your architecture decision is correct. They just work.

But AI has limits — judgment calls, domain expertise, regulatory knowledge, creative direction, customer relationships. The things that require a human who’s done this specific thing before.

A just-in-time team is built around those moments. A security auditor who plugs in for two weeks before launch — doing the work they’re best at, without sitting through six months of standups to get there. A compliance specialist for a regulatory filing. A designer for a branding sprint. A domain expert who validates your caregiving tools actually work for caregivers.

These aren’t interchangeable parts. They’re independent experts who choose this model because it’s better for them too. No corporate politics. No busywork to justify headcount. No slow death in a role that outgrew its purpose. They do their best work, on their terms, with people they respect.

This isn’t contracting — contracting is an employment workaround. And it’s not gig work — gig work commoditizes people. Just-in-time teams are mutual. The builder gets world-class expertise without permanent overhead. The expert gets meaningful work without organizational drag. Both sides choose to be there.

Why Now

Three things converged:

AI agents crossed the capability threshold. Not the “impressive demo” threshold — the “I actually ship production software with these every day” threshold. I run multiple autonomous AI agents across multiple machines, 24/7. They write code, run tests, manage builds. The baseline work is handled.

Remote work normalized human flexibility. Five years ago, getting a security expert for two weeks meant flying someone in or hiring a local consultancy. Now it means a Zoom call and a shared repo. The friction of just-in-time human expertise collapsed.

Large companies proved the counter-model fails. Every major tech company hired aggressively in 2021, then spent 2023-2025 laying off the same people. The permanent headcount model doesn’t match how work actually flows. It never did — companies just couldn’t see the alternative.

What This Looks Like in Practice

At Graham Alembic, the team is lean — mostly AI agents. I handle judgment, direction, and the things that require being human. When I need specialized expertise, it’s a partnership, not a transaction. People doing what they’re great at, exactly when it matters.

This isn’t about being cheap. It’s about being honest about what serves everyone best. A brilliant security engineer doesn’t want to sit idle for 11 months waiting for the two weeks a year their skills are critical. They want to do security work — for multiple teams, on hard problems, with autonomy. JIT teams give them that.

The result is a company that can build across multiple products — Claudine, Kindling, Alembic Compute — without the overhead that would make multi-product development impossible for a small operation. And every expert who works with us gets to do the thing they’re actually good at, not the thing that justifies a full-time seat.

The Network Effect

Here’s the part that gets interesting. Just-in-time teams aren’t just internal. They work between companies too.

When independent builders each have AI agents that can coordinate, collaboration becomes natural. You don’t need to merge companies or form partnerships with legal overhead. You need a shared repo and agents that know how to work together.

The relationships between builders become more like friendships between peers than dependencies in an org chart. You collaborate because the work calls for it, not because HR put you on the same team. The human connection is real precisely because it’s voluntary.

I think the future looks like networks of lean operations, each running on AI, with experts moving fluidly between them — contributing where they’re needed, building real relationships along the way. More fluid than corporate. More respectful than contractor marketplaces. The music never stopped — the band just got a lot more flexible about who sits in.

The Bet

Large companies are entering structural decline. Every permanent employee is overhead that AI is learning to replace. The economics are brutal and accelerating.

But lean teams with just-in-time human expertise? They get more capable as AI improves. Every advancement in AI agents makes the baseline stronger. The experts who plug in get to focus on higher-value work because the routine is handled. Everyone’s time is respected.

This is the model Graham Alembic is built on. Not because it’s trendy — because I’ve been living it, and it works. Every day. And the people I’ve worked with this way seem to prefer it too. It turns out most experts would rather do expert work than attend all-hands meetings.

If you’re building this way, or thinking about it — whether as a builder or as an independent expert — I’d like to hear about it. The best ideas here are going to come from people actually doing the work, not from McKinsey decks about “the future of work.”

Shall we go?