Two Z4s, One AI Race Engineer

Part 4 of my AI-coached track weekends. The loop that Barber proved with a coach and VIR proved without one gets its weirdest test yet: two cars, one weekend, and a repair job that ended nine days before it started.

Published August 19, 2026 — five weeks after the weekend it describes, and after Barber. That delay is a story of its own, and it's near the bottom.

The $12 part that planned my weekend

Seven days before Road Atlanta, I snapped a bleeder valve.

Not doing anything heroic — I was adding high-temp brake fluid to my 2008 Z4 Coupe, deliberately the only thing I wanted to change on the car. The plan was a baseline weekend: stock brakes, stock rotors, fresh track tires, and one concession to reality, because I knew stock fluid wouldn't survive Road Atlanta. The front-right bleeder had other ideas. It sheared off in the caliper, and just like that I had a track car with three working corners and a week to fix it.

But let me back up two weeks, because the reason I was this committed to the Coupe starts at VIR.

VIR: the proof of concept

At VIR I ran my M440i — heavier than anything you'd call a track special, and speed-limited to 130 mph. Between every session I uploaded my laps and worked with the AI on three things: braking g, minimum corner speeds, and where I wasn't committing. It had my whole history at that track to compare against — the M5, the old M340i, last year's Z4 M40i — so "where am I leaving time" wasn't a feeling, it was a diff.

The uphill Esses were the headline. The data showed I was breathing where the car didn't need me to, and fixing that one habit was worth most of a second by itself. Over the weekend the loop took me to a 2:13.14 — three and a half seconds under the 2:16.61 I'd set in the lighter Z4 M40i the year before, and 0.22 seconds off my all-time VIR best in the M5, a car the M440i has no business being near. The M5's straight-line speed is untouchable; the corners, it turns out, were negotiable.

That's why the Coupe mattered. My rule is: don't take the same car with the same setup back to the same track. The M440i had just done Road Atlanta. VIR proved the AI loop travels between cars. The E86 was next.

The thrash

So: snapped bleeder, seven days. If I was replacing a caliper anyway, I was replacing both fronts. If I was buying pads, they might as well be race pads. Rush shipping had new front calipers, four stock rotors, and Hawk DTC-60s on my doorstep by Wednesday.

Wednesday night, Thursday night, Friday night until ten: all four rotors, both front calipers, the full high-temp flush, and a bleed sequence I now know better than I ever wanted to. (The AI wrenched alongside me — torque specs, bleed order, and one honest sequence correction I'll spare you — the full garage-night story is its own post.) Meanwhile, as insurance, I'd already pulled the wheels and track tires off the M440i — the RE-71RZs that went on new before VIR, still fresh — and bolted them onto my other Z4, the 2022 G29 M40i. If the Coupe wasn't ready, the Roadster would go.

The Coupe was ready. And then I looked at two track-prepped Z4s sitting in my garage and thought: why choose?

Two cars, one paddock, fifteen minutes to analysis

Saturday I drove the Coupe up early. Meredith drove up at lunch and took me home, and the Coupe stayed at the track overnight. Sunday morning I arrived in the Roadster — and suddenly I had a 2008 Z4 and a 2022 Z4 in the same paddock, fourteen years and one whole philosophy of sports car apart. I moved my number stickers from one car to the other between run groups. One car cooled while the other ran.

Fourteen years apart, same paddock, same number stickers.

The loop worked the same as VIR: laps uploaded within fifteen minutes of each session, then a conversation about where I wasn't committing — braking, minimum speeds, the same three questions, now with two very different answers depending on which seat I'd just left.

Saturday, the Coupe's first track day ever — and my first serious manual work in years. I was learning heel-toe rev-matching on track, at speed, in a car that redlines at 7,000 and expects you to be there. The AI called my braking the soft spot after the first session (-0.4g where the tires had 1.0 available), and the day went 1:56 → 1:53 → 1:50.10.

Sunday, the Roadster — and the number I didn't know about. The session-by-session loop had me under my old personal best in three separate sessions. The best of them, a 1:40.53, was sitting in a midday session I almost didn't upload; the AI found it in the data two days later, verified it against its neighbors, and threw out three phantom laps while it was at it. December's PB was 1:43.26 — a 2.73-second gain, and not from the corner I expected.

Zone	December (1:43.26)	July (1:40.53)	What changed
T5	Distinct -0.58g brake zone (156→92 mph)	Never crosses -0.25g — a light lift, ~10 mph more carried	Brake event eliminated — a line/commitment change, not braking harder
10a/b → T12 complex	Brakes later (onset ~3,320m), harder (-0.76g peak)	Brakes earlier (~3,252m), softer (-0.69g peak)	Exits 10a/b at 53.0 mph vs 42.7 mph, carries 91.6 vs 78.6 mph through T12 — +13 mph through the final corner
T1 / T3 / T6-7	—	Modestly harder braking (-0.76 / -0.79 / -0.88g)	Smaller contributors

Look at that T5 row again, because it has a backstory. In June, at this same track, Levi — the chief instructor — told me I was braking for T5 too early, just before the rumble strips on the right, and to brake as far past them as the car would hold. I got one afternoon of practice on cooked tires. Five weeks later, on the RE-71RZs, "later" had become "the brake event is gone" — a light lift where a -0.58g stop used to live, ten more mph carried through the corner. And I spent the same weekend passing the exact same tip down to my own student: brake later into T5, well past the strips, if it holds. Coach to junior-coach to student, one braking point at a time.

The pre-session briefing said brake later and harder into 10a/b. The stopwatch said the opposite worked better there — brake earlier and softer, and let the earlier, smoother application preserve rotation all the way to T12's exit. Both were right, just about different parts of the same corner: the briefing's instinct was correct for T5 (eliminate the brake, carry the speed), and wrong for 10a/b, where carrying speed through the braking zone itself — not stabbing it — is what unlocked T12. T12's minimum speed is still 3-7 mph off where the target says it should be, so there's more in that corner yet.

The 1:40.53, onboard.

Sunday afternoon the Coupe came back out and finished its debut weekend at 1:48.99 — seven seconds faster than its first lap on Saturday, on stock camber, with brakes that were parts in a box five days earlier.

The Coupe's 1:48.99, onboard.

What the sensors saw

All three sessions ran at roughly 25Hz GPS+IMU, 166,000-plus fixes for the weekend once the Saturday upload finished recovering from a mid-handoff stall. Saturday's E86 debut ran GPS and g-forces only — no OBD in the car yet. By Sunday the OBDLink CX had moved over, so the M40i's PB session came back with RPM, throttle, coolant, oil, and intake-air temp on top of the position data, and the E86's afternoon session picked up RPM, throttle, and coolant when the dongle followed it back.

The numbers worth keeping: the M40i touched 145.5 mph, 1.29g lateral, and 6,944 rpm, with coolant peaking at 106°C and oil at 123°C in roughly 90°F ambient — the B58 ran hot all day and never complained. The E86, on the other hand, used every bit of the N52's 7,000-rpm redline (7,055 rpm recorded) with coolant up to 109°C in July traffic, which is worth watching on a car this old. The E86's stock-camber lateral grip came in at 1.15g for the debut day, and the soft-braking note from the first session stands as the biggest open item. One raw number to ignore: Saturday's file shows a -1.83g brake spike, which is a curb/noise artifact, not a real stop — the per-corner braking analysis above, capped at -0.90g into 10a/b, is the one to trust.

Still missing from the setup: brake-pedal percentage, steering angle, wheel speed, and fuel — all queued behind figuring out the right capture rig for a 2008 car (that's its own saga, and it ends with a physical oil-temp sender rather than more software — still on the list, not yet in the car).

Scoreboard

Car	Track	Before	After
M440i (VIR, June)	vs its own weekend	2:14.12 day 1	2:13.14
M440i (VIR)	vs the lighter Z4's 2025 best	2:16.61	-3.47s
M440i (VIR)	vs the M5 all-time	2:12.92	0.22s short
Z4 M40i (RA, July)	vs Dec PB 1:43.26		1:40.53
Z4 Coupe (RA debut)	first lap Saturday	~1:56	1:48.99

A snapped bleeder valve turned into a two-car weekend. An AI race engineer turned two new-to-the-track cars into personal bests. And a guy who'd driven automatics for years spent Saturday learning heel-toe on the one racetrack he knows best — which, honestly, might be the real story here.

The post that sat in drafts for five weeks

Time to eat some of this. The draft you just read was finished on July 13. You're reading it on August 19 — after Barber, the event the closing section below was written to preview.

So I ran the delay the way I'd run a bad session:

• The draft was edit-complete on July 13. Everything left was mine: a one-click publish flip, a header photo, and one open content question (where the storage-unit-full-of-tires story belongs).
• All three lived on one ticket. So the one-click item waited on the open question. The photo landed in the repo on July 17 — the exact paddock shot the ticket asked for, a foot from the finish line — and nobody went back to look.
• Nothing had a due date. "Within four weeks, before the next Barber" is a vibe, not a deadline, and no check was watching for it.
• Then the X5's AC died the week before Barber, the tow collapsed, the weekend swallowed everything, and I wrote and shipped the Barber post first, because Barber was the thing in front of me.

One AI race engineer, fifteen-minute session turnarounds, and it still let a finished post rot in drafts for five weeks. The tell is that I never asked it. It read every lap I handed it and nothing I didn't.

What changed: one post per event, published within a week of the event — enforced now, and no longer queued behind this one. Ship-it steps get split from open content questions, so a decision can't hold a publish hostage. And a standing sweep walks every open ticket and every unpublished draft asking what's gone stale. That sweep is what found this post.

Next up

The Coupe was supposed to go to Barber. It didn't. The week before the event the X5's front blower quit — no AC, in an Alabama August, with a car on a trailer behind me. The X5 went to my shop instead of to Leeds, and the E86 stayed in the garage.

So the Roadster went, and it worked out perfectly: still on the RE-71RZs, still set up exactly as it finished Road Atlanta, sitting there as the ideal backup track car. What happened once we got there is its own post — half a second between two completely different cars. I did feel a little bad for Smurfette after that weekend. She used to be the track car, and there she was, the backup.

Except she isn't, quite yet. Next on the calendar is Ozarks with Chin, September 5-6, and I went back and forth on which Z4 goes on the trailer. Learning heel-toe is one thing at Road Atlanta or Barber, where I know every inch of the place. Ozarks is long, with a pile of corners I've never seen — and a new track plus a gearbox I'm still learning is two unknowns at once. So after our first pass of Ozarks track prep and analysis, the call is made: Smurfette goes to Ozarks, on the back of a ShiftR trailer we're renting, and the E86 Coupe goes to AMP with JTI on September 12-13. Good news for Smurfette — un-demoted for a weekend. She'll take it. 😉

Before the Coupe's turn it gets a coolant-level check, because that 109°C number earned it a look. The proper oil-temp sender is still on the list and not in the car — the N52 won't tell me its oil temperature any other way, so it's coming, just not for this one. New track for the Roadster, new track for the Coupe, one instrument still missing. The loop continues.

Want your laps analyzed like this? Email your exports — Harry's LapTimer .hlptrl, or CSVs from Porsche PTPA, AiM, RaceBox — to upload@trackdays.ai.