Tom Stoffer

🧠 Director of Dev | 🎛️ DJ tech | 🏗️ Builder of products, systems & strategy
Chasing leverage in automation, markets & ideas. 🤖📈💡
#NZTech #AucklandAI

[Tom Stoffer @Stoff81]

Thanks to @steipete for implementing my feature request into Lobster. https://t.co/oMEzy31ZJ7

That’s gonna make my bookmark review and other workflows way snappier. No more handoff lag between cron lobster -> heartbeat -> cron lobster 🦞 ✅

Hopefully don’t have to wait too long before it lands in OpenClaw

[Tom Stoffer @Stoff81]

@steipete @irl_danB Thanks for the feature! Can’t wait for the next release to check it out

[Tom Stoffer @Stoff81]

@MatSilverstein What sorts of things have you found to be sticky? I'm building something similar for the New Zealand market

[Tom Stoffer @Stoff81]

@MatSilverstein @tobi @obisidan Just installed, looks great!

[Tom Stoffer @Stoff81]

This is how I’ve worked around it for now. Does feel like a gap though. This adds quite a bit of latency waiting for the next heartbeat/cron cycle. If we could implement the agent call to use the lobster resume token that would make it synchronous.

What’s your thought on injection intelligence into a “dumb” flow, seems backwards but maybe not?

Alternative would be to use prose for the workflow and have it call the lobster for state change.

[Tom Stoffer @Stoff81]

1/8

I shipped a bookmark → task pipeline today.

Save something on X. It ends up as an approved product spec and a task in your backlog. Fully automated except one approval step.

The hardest part wasn't the AI. It was getting the AI the right context.

🧵

[Peter Steinberger 🦞 @steipete]

@Stoff81 @irl_danB Interesting idea, Prs welcome!

[Tom Stoffer @Stoff81]

https://t.co/lJlG5CVacV

[Tom Stoffer @Stoff81]

1/8

I shipped a bookmark → task pipeline today.

Save something on X. It ends up as an approved product spec and a task in your backlog. Fully automated except one approval step.

The hardest part wasn't the AI. It was getting the AI the right context.

🧵

[Tom Stoffer @Stoff81]

4/8

Lobster doesn't have a native way to call into an agent runtime.

So I didn't make it.

Instead: Lobster advances items to `spec_requested` and stops. It's idempotent. It doesn't care when the work gets done.

[Tom Stoffer @Stoff81]

@zdogmode Read this

[Tom Stoffer @Stoff81]

1/8

I shipped a bookmark → task pipeline today.

Save something on X. It ends up as an approved product spec and a task in your backlog. Fully automated except one approval step.

The hardest part wasn't the AI. It was getting the AI the right context.

🧵

[Tom Stoffer @Stoff81]

@elponick This is my first refactor. I changed the schema considerably of the spec -> task creation. Are you on openclaw? What’s your control surface Telegram or other?

[Tom Stoffer @Stoff81]

@itshreyaagarwal Yes, worth the upfront investment

[Tom Stoffer @Stoff81]

Having taste is important for me, so I've been spending the week investing in some proper design tool. I've built up a design token library which exports to CSS, React Native and also a https://t.co/dzt1nkjoOE file.

The library exports to all these formats so I can ensure Web/iOS and Mockups never drift.

Next up will be syncing the primitives to a proper design system.

[Tom Stoffer @Stoff81]

Follow for a deep dive article coming soon…

[Tom Stoffer @Stoff81]

8/8

I keep asking: where does the intelligence need to live?

Some steps are mechanical — Lobster handles them.
Some need context that only accumulates in a running agent — Quinn handles them.

The art is drawing that line cleanly.

[Tom Stoffer @Stoff81]

@techxutkarsh Can you share with me pls