cd /news/ai-agents/your-rough-draft-is-someone-else-s-m… · home › topics › ai-agents › article
[ARTICLE · art-172696] src=sageox.ai ↗ pub= topic=ai-agents verified=true sentiment=↑ positive

Your rough draft is someone else's missing piece

SageOx's Ryan Snodgrass released a new version of the company's session protocol that streams every coding-agent turn to teammates in under a second, replacing the prior model where a session reached the team only after the developer exited Claude and the ox CLI rebuilt and committed it to the Ledger. The design combines "Twitter time" and "GitHub time," a framing attributed to Venkatesh Rao's Blygger protocol: each turn becomes a line stripped of secrets and sent through SageOx's API to S3, while the Ledger catches up minutes later with the versioned session folder. The post notes that pasted screenshots and videos now travel with sessions, uploaded to S3 and hash-checked, but warns that secret stripping reads text, not pixels, so users should crop before pasting.

key takeaway

SageOx's session protocol v2 emits each agent turn as a line to S3 in under a second, with the versioned Ledger record following a few minutes later, and secret stripping covers text only, so screenshots must be cropped before pasting.

by read7 min views1 publishedNov 18, 2026
Your rough draft is someone else's missing piece
Image: Sageox (auto-discovered)

Ideas from Blygger and Digital Gardens for 40x teams

The world knows little how many of the thoughts and theories which have passed through the minds of a scientific investigator have been crushed in silence and secrecy by his own severe criticism and adverse examinations.

Michael Faraday

In performance-blog-land you do that thinking and researching privately, then shove it out at the final moment. A grand flourish that hides the process.

After the Lisbon offsite, I stood up in front of the team today: We see how quickly ideas pile up. Small sparks end up as a group forest fire. The world outside is changing so fast that our group sense-making has to be updated as soon as possible. If any one of us is drunk on the status quo, then the right thing to do is a punch in the gut to get them unblocked.

On the same day, after this, Ryan announced a new version of our session protocol, where our sessions are going to be shared in real time.

In the past, the coding session used to reach the team at the end of a session. You would work with Claude for two hours, and the ox CLI would be writing every turn to a local file. And then at the end of your session, when you exit Claude, ox would rebuild the session and commit it to the Ledger. Two hours of your teammate's thinking, and you see none of it until they are done.

Twitter time and GitHub time #

One of the recent mental models that I learned was from Venkatesh Rao. He was talking about his new protocol called Blygger, and he talks about two kinds of time: Twitter time versus GitHub time.

Twitter time is the present moment. You post it, it scrolls away, the snapshot is the point. GitHub time is a sequence of states: versions, history, the changing is the point.

The git way of sharing works well for complicated problems which do well with logical decomposition. Not so well with wicked problems which do not have one magical solution that is going to come out of one genius's head: these complex problems need co-created solutions which involve the entire group's genius. What we actually riff with is our colleague's rough work.

The reason why Ryan and I are obsessed about real time is actually not speed. It is that when your rough draft shows up in front of the other person, the one who is holding the other half of the idea, something magical happens. It catches fire, and it changes from small sparks to a forest fire. And over time, this wonderful resonance allows a team to overcome its B.C. norms of perfectionism.

Session v2: Twitter time and GitHub time #

So in today's revision, what Ryan did was that he basically makes every turn a line, and that line leaves your laptop in under a second. We of course strip the secrets out of the text, and then these lines go through our API into S3.

And then what we do is very similar to Blygger: we combine Twitter time with GitHub time. The Ledger catches up with a sweep, a few minutes later, and publishes the session folder into the Ledger with the normal history and versions. A live path in Twitter time. A versioned record in GitHub time.

And one of the really interesting side effects is that suddenly images and videos come along. The screenshots you have been pasting into your sessions get shared with your teammates too. ox uploads them straight to S3, checked by their hash, and the session line points at them. One honest caveat: secret stripping reads text, not pixels, so crop before you paste.

So now you can watch all of this in both Twitter time and GitHub time, and you don't have to wait hours until your colleague's sessions are published.

The mirror image: speech over MoQ #

This is the mirror image of the work that I did a couple of months ago, when I wrote about how we stream audio and speech with MoQ. One of the very opinionated choices we made was to choose MoQ and not WebRTC. That argument, which I picked up from

Luke Curley, comes in five parts:

  1. WebRTC is built to drop. "WebRTC aggressively drops audio packets to keep latency low." Fine for two humans on a call. Wrong when the listener is a model or an archive you promised never to lose.
  2. The last packet is the one that matters. A model cannot answer until it has the whole prompt, so there is time to retransmit."A voice agent should never ask the user to retransmit a prompt."
  3. Machines are faster than real time. They can catch up, and they can talk in bursts.
  4. Every listener gets its own latency budget. On one SageOx broadcast, the live transcript waits milliseconds, Voice ID waits seconds, and the archiver waits for everything.
  5. WebRTC scales like a phone call, not like a CDN.

So what we're doing is investing deeply in foundational technology, to make sure that our human-to-human and our human-to-AI-teammate conversations are of course becoming part of the SageOx context. But we are also making sure that they become part of the SageOx context in real time.

Getting over the fear of ridicule #

This stops us from having this idea that there is a need for perfectionism, and it is how the fear of ridicule gets overcome. Because almost everything that we talk about is shared with the team in real time.

And then what happens is, when you're talking about a deep problem, especially during the kind of magical hackathon that we had in Lisbon, you can start riffing off each other's ideas, and get a sense of where the other person is at, even without them knowing exactly what you were doing.

I think and talk a lot about the incentive systems that would make people get over their fear of ridicule. And I realized that I'm walking down a path that economists of science have walked before. Science runs on credit, which goes to whoever publishes first. Paula Stephan puts it best: "the only way to make it yours is to give it away." The problem is that you want to have the acknowledgement for what you do. And so the same race that makes scientists publish makes them sit on materials and half-finished work. Even the money waits for the finished thing. At the NIH, it was common to say "no crystal, no grant." In structural biology, the hard part is growing the protein crystal you need before X-ray crystallography can show its shape. Reviewers wanted to see the crystal before they funded the work to get it. Hoard in private, release in GitHub time. Faraday's crushed ideas are partly a product of that incentive.

This is an incentive mechanic that just doesn't work in a world which is moving at 40x speed. And you can see that the communities of people working where the rate of change is the highest, like ML, have gone around the system by working on arXiv and Twitter. The credit is still there. It just runs at Twitter time.

Designed serendipity, as a matter of course #

By now, we see it clearly. Human attention is the most important resource we have, and right now it is getting whiplashed around and very inefficiently allocated. Every day there is a new model, a new tool, a new way of working, and everybody is trying to keep up on their own.

If everybody in the team gets to see each other's rough, dirty, small ideas, the entire team gets to make its way through the first, second and third order implications way faster. One person sees the idea, the next person sees what it breaks, the third person sees what it makes possible. Nobody has to do all of that thinking alone in their head. All of this only works if there is no fear of ridicule.

And I will make an even bolder claim. 40x teams are only possible when ideas are shared internally in Twitter time, instead of the B.C. GitHub time.

Your rough draft is someone else's missing piece. Let them see it.

── more in #ai-agents 4 stories · sorted by recency
── more on @sageox 3 stories trending now
sponsored brought to you by zahid.host 4,200+ EU-deployed projects
reading about agents? ship yours in a single git push.

Run your AI side-project on zahid.host

EU-based hosting, git-push deploys, automatic HTTPS, no cold starts. Free tier with a custom domain — perfect for shipping the agent you just read about.

$git push zahid main
→ Live at https://your-agent.zahid.host ✓
Get free account → Pricing
from €0/mo · no card required
LIVE [news/your-rough-draft-is-…] indexed:0 read:7min 2026-11-18 · —