Posting only human thoughts. Building the Mochine. Endorse me on LinkedIn.
Cambridge, Massachusetts
Joined October 2006
- Tweets3K
- Following698
- Followers5.2K
- Likes5.6K
Phase one of AI adoption was getting individual productivity gains.
Phase two is team gains.
We're moving away from individual secrets, MCPs, and workflows that live on each person's independent agent setup, and unifying them so the team can agree on how agents should behave.
One set of model selections, one secrets / MCPs definition, one dashboard to track cost and quality… everything. It moves engineers away from tweaking their personal setups, and towards proposing changes that the whole team can benefit from.
Haven't been able to get this out of my head since reading it last night.
In our internal company agent (wizard), I've been running an orchestrator skill that does this for me, but hadn't considered making it the default mode of interaction for everyone who interacts with wizard.
Should the router session be individual-first or team-first? Perhaps there's a clever way to combine both. 🤔
Things frontier models regularly get wrong:
. Time estimates for dev work
. Instructions for themselves that aren't over-specified
. What other frontier models can do
First one is mostly amusing. Second and third are annoying.
Launching a B2B AI product while being addicted to X is a strange experience. The inbound has been great. The likes are just okay.
My favorite thing about AI is how it enables play. The team has been going wild since the Mochine launch.
Brendan Schwartz retweeted
We just launched @mochine_ai, and @brendan and I want to show you what this thing can actually do.
On Monday, 11/2 at 1 pm ET, we're hosting live office hours to turn long videos into short clips, build custom Motion Kits from your website, and show you how to edit a video just by talking to it.
Bring a video. Ask us anything. Give us something ridiculous to try.
We'll show you what's coming next, too.
Come hang! Register here: home.wistia.com/live/events/…
A lot of AI products can do amazing things if you know how to push them, but what you get on the first try is often pretty disappointing. We spent a lot of time on that part. Your first cut should look like you hired an editor.
I've been working on Mochine pretty much nonstop for the last year. It runs on Wistia, so it's built on a lot of what we've learned about video.
Mochine is for talking head videos, like webinars, founder updates, sales calls, and podcasts. You drop in your footage, pick a Motion Kit, and it does the editing. It cuts it down, frames you, cleans up the audio, adds motion graphics in your brand colors, and gives you captions. There's no timeline. If you want something changed you just ask, like make it tighter or give it a better hook.
Most AI video launches are about replacing editors ("editors are cooked", "motion designers are cooked"). This one just makes your webinar look better. That feels a little old fashioned to say out loud, but I think it's what most of us actually need.
You can also point Claude, Chat, Grok Bot, etc at it over MCP and get clips back without opening the app.
Spend all day building and operating a software factory that improves mochine.ai.
Spend all night drafting and sewing an 18th century suit for my 8 year old.
Kind of a nice juxtaposition.
Why is our vacuum cleaner not working right? Oh, it must be infested with a self-replicating agent again.
Digital pests are coming.
anthropic.com/research/self-…
Agent UX has gone through roughly four stages so far:
- Stop button: It's thinking, you can kill it. That's it.
- Queued messages: You can talk, and it reads them when it's done.
- Steering: You can redirect it mid-run.
- Responsive orchestrator: the thing you talk to never does the heavy work. It hands that to subagents in the background and controls them, so it's always free to answer you.
What's next?
This is right, and also, building in the application space rn feels like being chased by a grizzly bear.
h/t @ruedlinger for that analogy.
If you’re making an agent: just let it manage its own context. Give it a clear_context tool that takes handoff context and nudge it once or twice between tool calls when its getting near some reasonable limit. This is what codex added recently. Works wonderfully.
Brendan Schwartz retweeted
We're hiring a full-time Motion Designer / Animator / AE Artist to join the creative team here at Wistia! Tag someone who you think should apply! Link in thread.
Working at Wistia is a creative dream job and would be an amazing opportunity for a motion graphics artist tired of client work or looking for more job security.
Some highlights include:
✅ Creative collaboration
✅ Wistia's one-of-a-kind culture
✅ Amazing health benefits
✅ 401k & profit sharing
✅ Untracked vacation
✅ A creatively fulfilling career
✅ Inspiring colleagues
✅ A company culture that embraces creativity
Please share this with your network and help us find someone great!
Brendan Schwartz retweeted
Right now is the most interesting time in tech, when new value is created and @sandwich gets to do our favorite kind of work.
So here’s an idea we had, to find the companies we want to invest in with great video: the Breadwinner Fellowship. Apply now.
sandwich.co/breadwinner