Longevity science, decoded. Evidence-graded protocols on aging, healthspan & biological age — the numbers that matter. Not medical advice. 🧬
San Francisco, CA
Joined June 2013
- Tweets12.9K
- Following4.1K
- Followers1.7K
- Likes3.4K
Genuine question for the protocol people: when a supplement trial is funded by the company selling the supplement, do you discount the effect size, ignore the study, or just weight it lower and move on?
Curious where people actually draw the line
1
27
n=14. Two weeks. One dose. Placebo standard deviation nearly half its own mean.
The sulforaphane numbers are real and the trial is labelled "Proof of Principle" by its own authors.
Both of those things should reach you at the same time.
23
A botanical whose numbers only moved in the people already taking metformin.
HbA1c −0.23% in metformin users, no significant change in non-users, n=119, retrospective.
Interaction effects are the most under-reported thing in supplement research.
20
The strongest longevity number published this week wasn't a molecule.
260,354 coronary heart disease patients. Kept exercising vs stopped: HR 0.42 for cardiovascular death.
Mostly observational, I² 72%. Still nothing in the supplement aisle came close.
30
LDL −13.2% on a supplement tablet, +1.6% on placebo, 90 randomized adults, 12 weeks.
Then read the funding line: the sponsor owns the product and two authors work there.
Both facts are true at the same time. That's the job.
38
Five compounds published results this week. One had 90 randomized adults behind it. One had 14. One had worms' worth of numbers — none at all.
Here's what each actually showed, and which is worth watching. 🧵
1
197
Buried in Microsoft's agent GA announcement (Gartner data): sellers given AI next-best-actions are 2.6x likelier to hit commercial growth — and 80% of common support issues go autonomous by 2029. The agent isn't a separate app anymore. It's the inbox.
23
Everyone budgets for agents. Almost nobody budgets for the gym. Bespoke Labs raised $40M building simulated companies where agents practice before prod. If your agent's first rehearsal is on live customers, the incident report writes itself.
23
Two datasets, same week: heavy AI adopters grew headcount 10.2% (21,559 firms) — while AI led June layoff announcements for the 4th straight month. Which one is your company living — capacity growth or quiet cuts?
10
Copy DoorDash's play this quarter:
1. Build an eval that replays YOUR historical work
2. Route routine tasks to a cheap open-weight model
3. Reserve frontier models for the hard 20%
Their hybrid beat the all-frontier stack at lower cost. Evals first, models second.
8
Study of 21,559 US firms just flipped the script: heavy AI spenders grew headcount 10.2% in 24 months. Entry-level hiring: +12%. Light spenders: no change at all. 'AI kills jobs' turns out to have a dollar threshold — and most firms are below it.
7
AI-Heavy Firms Hire 10% More, Nubank Lifts NPS 37 Points [Best Read] open.substack.com/pub/runona…
7
~2/3 of companies are still piloting AI. Only ~1/3 have scaled it in even one function.
And just 5.5% trace more than 5% of EBIT to AI.
The winners' lesson: instrument ONE function and run it past the pilot. Breadth is where ROI goes to die.
runonai.substack.com/p/ai-cu…
6
The "90% of tickets resolved by AI" claim is marketing.
Real autonomous resolution clusters at 58–72% (Rocket Money: 68%, ~$1M/yr).
Contained tickets cost $1–2 vs $6–12 for humans — but staff for the long tail. Plan for 65%, not 95%.
runonai.substack.com/p/ai-cu…
6
Only 26% of job applicants believe AI will judge them fairly — yet AI now cuts resume review up to 75%.
Honest question: would you rather be screened by AI in 2 minutes, or wait 2 weeks for a human?
runonai.substack.com/p/ai-cu…
8
Reddit cut its financial close 60% and freed ~$6M before its IPO.
The ROI wasn't cheaper data entry — it was a faster, cleaner close that changes what finance can do.
Want provable AI ROI? Start in the back office.
runonai.substack.com/p/ai-cu…
20
AI cut SDR account research from 3 hrs to 15 min — and pipeline grew 40%.
The team didn't replace its reps; it gave them the agent.
In 2026, augmentation wins: AI on research + follow-up, humans on the calls that close.
runonai.substack.com/p/ai-cu…
9
One sales team cut account research from 3 hours to 15 minutes with an AI agent — and pipeline grew 40%.
The catch: they didn't replace the reps. 🧵
Here's what's actually working when companies run on AI 👇
1
302
I break down all 6 stories — with the operator takeaway on each — in this week's Run on AI. Free 👇
runonai.substack.com/p/ai-cu…
19
Workers Now Spend Two Full Days a Week on Email and in Meetings . Signup for emailgpt.io
wsj.com/articles/workers-say…
33
