credit burn
1 views9/18/2026
When AI adoption moves faster than enterprise decision-making
0 Comments
Be the first to leave a comment.
▸Show captions (49)
| 00:00 - 00:03 | September credit usage is high again. |
| 00:04 - 00:05 | But we have 75,000 credits. |
| 00:05 - 00:07 | Usage should stabilize. |
| 00:08 - 00:12 | Management has repeatedly discussed the situation. |
| 00:12 - 00:15 | Surely we can make it through September. |
| 00:17 - 00:19 | And the latest burn forecast? |
| 00:19 - 00:21 | Everything will be fine? |
| 00:24 - 00:26 | My Führer... |
| 00:27 - 00:28 | The model... |
| 00:31 - 00:33 | At the current burn rate... |
| 00:34 - 00:36 | ...we run out September 23. |
| 00:53 - 00:58 | Everyone leave except IT, Architecture, and whoever owns the AI strategy. |
| 01:13 - 01:15 | THIS IS THE THIRD TIME! |
| 01:15 - 01:17 | WE RAN OUT IN JUNE! |
| 01:18 - 01:23 | I WROTE AN INCIDENT REVIEW! NO MONITORING! NO OWNER! NO OVERAGE PLAN! |
| 01:25 - 01:28 | THEN JULY CAME! |
| 01:29 - 01:31 | WE PREDICTED DEPLETION! |
| 01:31 - 01:34 | AND WE STILL RAN OUT! |
| 01:34 - 01:37 | SO I BUILT MONTE CARLO SIMULATIONS! |
| 01:37 - 01:40 | BUSINESS DAYS! WEEKENDS! HOLIDAYS! |
| 01:40 - 01:42 | August actually had credits left— |
| 01:42 - 01:46 | BECAUSE I BEGGED EVERYONE TO PUMP THE BRAKES! |
| 01:46 - 01:48 | Users are choosing cheaper modes now— |
| 01:48 - 01:52 | I WROTE A 14-PAGE GUIDE ABOUT THAT! |
| 01:53 - 01:54 | INSTANT: ZERO! |
| 01:56 - 01:57 | THINKING: TEN! |
| 01:57 - 02:00 | PRO: FIFTY CREDITS PER MESSAGE! |
| 02:00 - 02:03 | I EVEN PUT A LUNCH TOKEN ON THE LINE! |
| 02:04 - 02:08 | AND AFTER ALL THAT, SEPTEMBER IS BURNING EVEN FASTER! |
| 02:08 - 02:13 | WE'RE ON TRACK TO RUN OUT A FULL WEEK BEFORE MONTH-END! |
| 02:14 - 02:16 | AND THE ENGINEERS WANT CODEX! |
| 02:17 - 02:21 | "CODEX AND THEN CODEX AND THEN MAYBE CODEX." |
| 02:27 - 02:29 | But the plugin is still "in progress." |
| 02:30 - 02:34 | We have usage data for a tool we haven't approved. |
| 02:34 - 02:36 | Fine. Build the AI gateway. |
| 02:41 - 02:42 | Authentication. |
| 02:43 - 02:47 | Logging. Cost controls. Multiple models. One controlled API layer. |
| 02:48 - 02:53 | Something we can actually integrate into our products! |
| 02:54 - 02:56 | How far have we gotten? |
| 02:56 - 02:59 | We discussed it again at Technology Meeting. |
| 03:00 - 03:02 | OF COURSE WE DID. |
| 03:04 - 03:07 | Maybe they'll ask him for another dashboard. |
| 03:14 - 03:16 | What more evidence do they need? |
| 03:19 - 03:23 | Usage reports. Heat maps. Gini coefficients. Burn forecasts. |
| 03:25 - 03:26 | Monte Carlo. |
| 03:31 - 03:33 | Employee productivity surveys. |
| 03:40 - 03:46 | I'm using AI to analyze AI so I can explain why we need to make decisions about AI. |
| 03:46 - 03:49 | And we still don't have Codex. |
| 03:53 - 03:56 | Fine, we'll do a performance review |