ByteDance Just Told Its Managers:
Instructions Aren’t Output.

What ByteDance actually did

In late June 2026, ByteDance CEO Liang Rubo sent an email to the entire company — his first all-staff memo in about four years. He rewrote the company’s “10 Leadership Principles,” and buried in it was a demand aimed straight at middle and senior managers: they can no longer run their teams by reading reports and sitting through briefings. They have to go into the frontline — the actual product, the actual users, the actual problems — and see it first-hand.

And this isn’t a motivational poster. The new standards are wired directly into the first-half 2026 performance reviews and tied to promotions and role changes. The Chinese tech press frames it as ByteDance attacking “big-company disease” — the accumulated layers of managers who coordinate, relay and align but never touch the work — as it retools the whole organisation for the AI era. A manager who used to coast on reports and past wins now has to show up where the work happens, or get re-rated.

Giving instructions used to be the output

For decades, issuing instructions was a real contribution. The manager was the person who had seen it before, held the whole plan in their head, decided who does what, and turned years of experience into direction. That coordination and judgment was genuinely valuable — it was the reason the layer existed and got paid. If you’d sat through ten product cycles, your instinct was the asset, and “I know what to do here” was worth a salary.

So when a company suddenly says “instructions aren’t enough, show me the work,” the honest question is: what changed? Managers didn’t get lazier this year. Something moved underneath them.

AI quietly took the part that made instructions valuable

Here’s my argument, and it’s the uncomfortable part. The specific things that made a manager’s instructions valuable — synthesising a messy situation, drafting the plan, anticipating the second-order problem, pulling the right reference from memory — are exactly the things AI now does in seconds, for anyone. When the planning layer is a prompt away, “I have the experience to tell you what to do” stops being scarce. In my own work, AI drafts a plan faster than any manager working from memory, and often a better one.

Experience hasn’t become worthless. But the part of it you used to sell — the plan, the instruction, the coordination — just got commoditised. What’s left that AI cannot hand you is contact with reality: what the customer actually did, what broke on the floor, the thing the dashboard doesn’t show. That is precisely why ByteDance is pushing managers to the frontline. The frontline is the one input AI can’t fetch for you — so it’s the one place a manager can still add something a model can’t.

This isn’t a ByteDance story

Don’t read this as one Chinese company’s quirk. ByteDance is just early, and blunt enough to write it down. The logic applies to every company with a management layer: when AI absorbs the analysis and the planning, the market re-prices what a manager is actually for.

I wrote about this from the shareholder’s side a few weeks ago — what are we actually paying senior management for — and the answer kept shrinking to two things: accountability, and genuine contact with the work. ByteDance just took the second one and made it a KPI. That’s the whole move: the reports and the slide decks are now a free feature of everyone’s AI, so the company stops paying for them and starts paying for the part that isn’t.

I learned it from zero. My team said they “used AI.”

Let me make this personal, because I’ve lived the exact split ByteDance is describing. When AI arrived, I didn’t delegate it — I sat down and learned it from zero, myself. Before that, I did try the normal thing: I asked my team to implement AI for the business. What came back was “oh yes, I use AI to enhance my videos.” That was the whole answer. Touching up a clip with a tool is not implementing AI, and the distance between those two things is the entire point of this article. So I stopped waiting and moved myself — into the prompts, the workflows, the things that broke.

Here’s the question I keep coming back to. If I, as the CEO, had to get on the ground to actually understand AI, why does a manager who says he wants to learn it still sit in his chair and only give instructions? That isn’t a skills gap — skills you can teach in a weekend. It’s an attitude gap. And it’s the thing I keep flagging about too much of middle management: they still feel they sit above the executive doing the hands-on work, they still believe their job is to instruct and be reported to. In the old world you could carry that posture for years and no one noticed. In the AI era it’s the fastest way to become the person whose cost the company can no longer explain — because the CEO went to the ground, and they wouldn’t.

If you manage people, read this twice

If you manage people — and I do — the instruction is simple and a little scary. Your value is no longer the instructions you issue or the years on your badge. It’s measurable output and first-hand contact with the work. Use AI for the planning; it’s better at that than your memory. Then spend the hours you save on the frontline, where the signal AI can’t reach still lives.

The manager who only forwards, relays and “aligns” is now the most exposed person in the building — because that is the exact job a model does for free. The one who ships something real and stays close to the customer is the one AI makes stronger, not redundant. Same title, opposite fate, and the thing that separates them is output.

The bottom line

ByteDance didn’t send that memo because its managers stopped working. It sent it because the definition of the work changed underneath them. When a machine can plan, giving instructions isn’t a contribution — output is. Experience still matters, but only the part that touches reality; the part that lived in reports and decks is now something anyone’s AI produces on demand. Get to the frontline and ship, or get re-priced.

Published by IMA AI — July 2026. Written by an operator who manages people and uses AI to plan every day — from inside the change, not above it.