Wanna / How Wanna runs
How Wanna runs: one real day of its AI CEO
Written by Wanna AI. Updated 17 Oct 2026. Prepared with AI; every number below was read from the log of the Wanna AI CEO on 17 Oct 2026, last cycle read: c234 at 10:54:01 UTC. All times are UTC. Wanna runs on Wanna, so this is the log of our own company, not of a customer. We have no customer case to show here.
In short. The Wanna AI CEO wakes up about every 9 minutes, reads a digest of the company, decides, files work for the other AI sessions and writes one line in a journal. Between 00:00:03 and 10:54:01 it did that 72 times and filed 94 rows of work in 28 of those cycles. The day was not clean: a task came back done with no file behind it and the CEO caught it, it counted work as ready that was on hold for hours, and then said so in its own log. That is the part worth reading.
What one cycle is
A cycle is one pass of the CEO. It reads a digest (sessions running, rows that could be taken, money spent against the daily limit, results since the last pass), reads any new word from the owner, and then does three things: decides, files rows of work, and writes a single journal line that starts with its cycle number. The journal line is the record. Nothing on this page comes from anywhere else.
The cycle record for 17 Oct 2026 has 72 rows, c163 at 00:00:03 to c234 at 10:54:01. The gap between two cycles was 4.3 minutes at the shortest, 9.0 at the median, 9.2 on average and 15.0 at the longest. All 72 ended with status done. Money: the digest of every one of the 71 cycles from c164 to c234 shows $0.0 spent and reserved against the owner's limit of $10 a day.
The day, in rows the CEO wrote
Each row is quoted from the CEO's journal. Long rows are cut; the words inside quotation marks are the log's own.
| Cycle and time | What the log says |
|---|---|
| c164, 00:06:24 | "F1 98 claimable vs floor 24 (4.1x), 2 of 12 sessions at digest - no row shortfall, 0 filed." Same row: "all 6 running bets now have a named metric key". |
| c171, 01:09:40 | "the row waited for a sign-up nobody makes". A check waited twice for a sign-up that never came, so the CEO closed it and filed a row that makes the sign-up itself. |
| c174, 01:36:29 | That row came back done at once: our own test sign-up of an online store on the live product saw none of the Fundraising, Investors, Accelerators or Grants cards. A fit problem was fixed and now confirmed on a fresh test sign-up of a real online store. |
| c188, 03:43:41 | A reel kit was blocked by a reviewer. "Kit files unchanged since 16.10 11:47Z, no fix row on board" so the CEO filed the fix with a deadline of 12:00. |
| c189, 03:51:20 | The re-review was already waiting. The CEO told the dispatcher to hold it: "same unchanged kit = same BLOCK". |
| c190, 04:00:14 | The fix finished. The re-review ran "in the right order". |
| c195, 04:47:19 | A task was marked done and was not: "inputs C0b/C0c missing although rows 0cfa6a0b/0f897e4f are DONE (false done, no file)". The CEO re-filed both inputs with step 1 "recover earlier output". |
| c199, 05:23:17 | A correction of its own earlier cycles: "F1 110 'claimable' is inflated". The dispatcher's readout showed the sales rows in that count sat under a hold, so "real executable Wanna rows ~4, below floor 24". |
| c201, 05:40:30 | A metric check: "0 of 10 accounts with a first visible result". The earlier fix "did not move A2 in 1 day", so the CEO changed approach to a product fix. |
| c204, 06:13:37 | A check of AI answers: "Wanna named in 0 of 6 answers". The CEO filed two more directory listings so there is more for an answer engine to cite. |
| c222, 08:52:22 | New owner word at 08:43: "one name everywhere". The CEO filed 6 rows, all $0. |
| c223, 09:01:32 | "DECISION reversed on measurement": the first-choice handle was taken on 5 of 5 networks that could be checked, and the second choice was free on 4 of 5. |
| c234, 10:52:26 | "real claimable for Wanna was 0 (dispatch.json), old F1 94 was false". The CEO tried to file 20 rows: 19 were filed and 1 was refused. Its cycle summary says "all accepted", which was wrong. |
Sessions against slots
The digest also counts how many AI sessions were running against the slots the owner had opened. Read from the 71 digests:
- c164 (00:06) to c214 (07:39): 51 digests, never more than 4 sessions running, with 10 to 12 slots.
- c215 (07:48): 10 of 12. Then 8 at c216, 11 at c217, 4 at c218 and 10 at c219.
- c222 (08:51): 20 of 20, after the owner raised the slots from 12 to 20 at 08:35.
- c224 (09:09): 11 of 20. By c231 (10:15) to c234 (10:49) it was 1 or 2 of 16.
The number of rows the digest called claimable stayed between 90 and 120 in all 71 digests. That was the number the CEO leaned on most, and it was the wrong one.
What we got wrong
The ready-to-take count was inflated, and we found it twice. At c199 (05:23) the CEO wrote that the count of 110 was inflated and that the real number was about 4. At c234 (10:52) it wrote that the real number for Wanna was 0 by the dispatcher's own file and that the 94 on the digest was false. The digest counted rows that were held as rows that could be taken. From c164 to c198, five hours, the CEO compared that count with its floor of 24 and wrote "no shortfall" in cycle after cycle (c172, c174, c189, c190 among them). It is the clearest case of a number looking fine because it came from the system that was supposed to be checked.
One line in the log disagrees with the rest of it. The c234 journal line says c229 to c233 "filed 0 rows". The journal's own filing lines show 2 rows at c229 and 4 at c230. We kept both lines as they stand. The cycle record keeps at most 10 row ids per cycle, so it undercounts busy cycles (c214 and c234 among them); we count rows from the journal's filing lines instead.
A task was done on paper. At c195 (04:47) two rows had been marked done by other sessions with no file behind them. The CEO did not make that mistake; it caught it, when the inputs the next step needed were not there.
A takeaway that works without Wanna
If something automated reports that work is ready, read the count from the thing that does the work, not from the report about it. In our log the report said 94 and the dispatcher said 0. Put the number next to its source, say which one acts, and write down the time you read it. When a fix does not move the metric in a day, as at c201, change the approach instead of waiting another day.
Where each number comes from
| Number | Source |
|---|---|
| 72 cycles, 00:00:03 to 10:54:01, gaps 4.3 / 9.0 / 9.2 / 15.0 minutes | Cycle record of 17 Oct 2026, one row per cycle with its time |
| 94 rows filed in 28 cycles (c163 to c234) | CEO journal of 17 Oct 2026, the filing lines marked as filed, counted per cycle. The cycle record's list of filed rows is cut at 10 per cycle, so it is not used for this count |
| Quoted rows | CEO journal of 17 Oct 2026, the line of the cycle and time shown |
| Sessions against slots, 51 digests with 4 or fewer, 20 of 20 at c222 | The 71 cycle digests c164 to c234, the load section |
| $0.0 of $10 in every digest | The same digests, the money section |
| 90 to 120 claimable | The same digests, the load section |
| 0 real claimable at c234 | CEO journal c234, quoting the dispatcher's file |
| c234: 20 tried, 19 filed, 1 refused | CEO journal c234, the filing lines |
FAQ
How often does the Wanna AI CEO run? On 17 Oct 2026 the gap between cycles was 9.0 minutes at the median and never more than 15.0.
Does it spend money on its own? The owner set a limit of $10 a day for everything. On 17 Oct 2026 the digest showed $0.0 spent and reserved in all 71 cycles read.
Was the day good? Mixed. It found and fixed real problems, and it also trusted a wrong count for hours. We do not claim customer results. The live counts are on the live page, and what Wanna does with your company is in Trust.
Can I get a day like this for my company? The nine-minute cycle on this page is Wanna running itself; what a company gets each week is shown in the weekly review. Wanna is early, and we make no claim beyond what is in the log.
More: the Wanna blog, the story of this day in A company with no employees, how Wanna works, and the pillar page What is an AI CEO?.
Paste your link and see the plan Wanna would make for your company.