Tesla is not waiting for the reveal. On its Q1 2026 earnings call, Elon Musk said Optimus production would begin at Fremont in late July or August 2026 — roughly four months after the last Model S and Model X rolled off that same line in early May. The car line is being torn down and re-tooled to build a humanoid the company has still not publicly shown. Electrek reported on 22 April 2026 that the reveal had slipped again; the original target was a Q1 unveil. So read the order of events carefully: the factory moved, the robot did not. For a desk like mine — where the whole job is the gap between a demo and a shift on a real floor — that inversion is the headline.

Here is why it lands. For a decade the hard part of a humanoid looked like the brain: the VLA policy, the balance controller, the hand that doesn’t crush the egg. Those problems are far from solved, but they now demo well. What does not demo is a line that can build the thing at rate. Musk put it plainly on X on 1 July 2026: “Optimus production will be extremely slow at first, as everything is new. This is not like making a car.” He cited roughly 10,000 unique parts on an entirely new line, called this year’s rate “literally impossible to predict,” and declined to give a 2026 target at all. The circulated V3 spec sheet comes only from aggregator blogs, not from Tesla; treat it as unverified. The confirmed number that matters is the parts count, because that is what sets the line.

The labour seam is worth naming, because it is quieter than the marketing. “Simple skills in the factory” means the first Optimus shift is inside the plant that builds Optimus — the machine’s first job is to help make more of itself. In January 2026 Musk acknowledged no Optimus robots were yet doing useful work at Tesla. That is not a knock; it is the honest shape of a new manufacturing curve.

←TODAY: Tesla is converting a Fremont car line to build a humanoid it has not yet revealed — the tooling is ahead of the robot. →3012: The android on the Zurich site is unremarkable, and nobody remembers which demo video it came from — only which supplier kept the actuators flowing. Fulcrum: The bottleneck in humanoids has already left the stage and moved to the loading dock; the demo was never the constraint.

Now the line into your practice. Construction wants humanoids — the lineage runs straight back through ETH’s robotic-fabrication work catalogued in The Robotic Touch, and ANYmal, the ETH Zürich spin-off, is already walking real European sites. But what decides whether a humanoid ever reaches a Baustelle is not the demo reel. It is actuator supply, parts count and line rate. That is why the leading indicator to watch is the supplier tape, not the launch event: BYD confirmed its first humanoid at the end of July 2026, and tier-one suppliers are repositioning — HL Mando, for one, has pushed an actuator line explicitly aimed at Optimus-class robots. When the actuators get cheap and plentiful, the site robot follows. When they don’t, the best VLA in the world stays a video.

Hack: Turn Musk’s “extremely slow at first” into a number your procurement can actually hold. The trick is that annual output is set by the slowest binding station, not the flashiest one — line rate, not policy model. Estimate it from takt time in three lines of Python:

shifts, hrs, days = 2, 8, 250          # your assumptions, not Tesla's
seconds = shifts * hrs * 3600 * days   # available production seconds/year
takt = 900                             # sec at the binding station — tune this
print(round(seconds / takt))           # ~16,000 units/yr at 15-min takt

Change takt to 300 seconds and the number quadruples; that single station is the whole story. Ask any vendor pitching you a site robot what their binding-station takt is — if they answer with a demo instead of a number, you have your answer.

Atelier: A Büro living with AI this year does not need to trial a humanoid — none is shipping to sites yet. What it needs is a supplier-watch habit, because that is where the real signal is. This Monday, open a one-page sheet and, for every robot your office is evaluating, write down the name of the actuator vendor behind it and that part’s quoted lead time — that column, updated monthly, will tell you which robot is real long before any reveal does.

PAZ has covered the acceptance side of this thread before: our piece on the RAI Institute’s study, where 1,000 people drove Boston Dynamics’ Spot at a Boston mall, found that a robot can pass every technical benchmark and still be parked within a month if the people on the floor aren’t comfortable with it. Supply gets the machine to the door; comfort decides whether it stays. Decide who is accountable when it gets something wrong before you put a face on it — not after. Start the supplier column now.