Transparency earns trust, so we’re building in public. Follow along as we build tools and workflows that turn AI from a black box into shared knowledge.
| # | User | Team | Daily usage | Tokens |
|---|---|---|---|---|
| #1 | so🎙HOST | plow | Th 06-25: 1.9BFr 06-26: 2.1BSa 06-27: 1.4BSu 06-28: 2.6BMo 06-29: 2.0BTu 06-30: 2.5BWe 07-01: 1.8B | 12.2B |
| #2 | wesm🎙EP.01 | posit | Th 06-25: 980MFr 06-26: 1.3BSa 06-27: 1.9BSu 06-28: 620MMo 06-29: 1.6BTu 06-30: 1.2BWe 07-01: 900M | 7.6B |
| #3 | naveen🎙EP.03 | skywillow | Th 06-25: 1.5BFr 06-26: 880MSa 06-27: 540MSu 06-28: 1.1BMo 06-29: 1.7BTu 06-30: 850MWe 07-01: 520M | 6.4B |
| #4 | erans | canyonroad | Th 06-25: 480MFr 06-26: 1.0BSa 06-27: 720MSu 06-28: 1.3BMo 06-29: 260MTu 06-30: 690MWe 07-01: 950M | 4.9B |
| #5 | jpmarindiaz | datasketch | Th 06-25: 700MFr 06-26: 460MSa 06-27: 980MSu 06-28: 680MMo 06-29: 1.2BTu 06-30: 500MWe 07-01: 280M | 4.5B |
| #6 | aashig | default | Th 06-25: 240MFr 06-26: 720MSa 06-27: 990MSu 06-28: 470MMo 06-29: 700MTu 06-30: 1.2BWe 07-01: 500M | 4.2B |
| #7 | plucas | plow | Th 06-25: 660MFr 06-26: 440MSa 06-27: 220MSu 06-28: 950MMo 06-29: 680MTu 06-30: 920MWe 07-01: 480M | 4.0B |
| #8 | danedelattre | plow | Th 06-25: 430MFr 06-26: 680MSa 06-27: 210MSu 06-28: 660MMo 06-29: 910MTu 06-30: 450MWe 07-01: 640M | 3.8B |
| #9 | wan | onewill.ai | Th 06-25: 200MFr 06-26: 420MSa 06-27: 640MSu 06-28: 440MMo 06-29: 880MTu 06-30: 190MWe 07-01: 410M | 3.3B |
| #10 | 2bert | cncorp | Th 06-25: 180MFr 06-26: 380MSa 06-27: 160MSu 06-28: 620MMo 06-29: 400MTu 06-30: 170MWe 07-01: 360M | 2.4B |
Daily Yield is where the Plow team shares what we’re learning as we build. Every day at 2pm PST we go live to break down practical workflows, experiments, and lessons from the front lines of AI.
Every agent needs access to do real things: your email, your files, your money. The industry is mostly ignoring the hard part — how do you grant access granularly, revoke it cleanly, and audit what ran without making the whole thing unusable? Containers and per-container permissions are the current best answer. What does that actually look like in practice?
The tools work. Anyone can spin up something custom in an afternoon. But the moment you want to hand it to someone else, the questions multiply: is it safe to run on their machine, does it work in their context, can they trust where it came from? Distribution is the unsolved layer. That’s the gap Plow fills.
When software is free and everyone is installing seeds from strangers, the old “only download from trusted sources” heuristic collapses. NPM-style dependency attacks, prompt injection in instructions, malicious seeds that look legitimate: this is the attack surface Plow’s trust layer is designed to address.
Traditional software testing assumes repeatability. Agents introduce non-determinism at the execution layer. How do you QA a seed? How do you know if it worked? This is an unsolved and largely undiscussed problem that the Plow team is navigating in real time.
GitHub gave code a home: findable, forkable, with a trust signal baked in — stars, contributors, commit history. None of that exists for seeds yet. That’s the gap Plow is building into. What does the GitHub moment look like when the unit isn’t a repo, it’s an instruction file?
World-class AI practitioners show exactly how they work. No predictions, no hype — just workflow. Hosted by Sam Odio & Tom Preston-Werner. Guests come straight off the leaderboard.
Apply to be a guest ⟶
LATEST



