Chris Bell | knock.app

We added an AI step to our workflow engine to help teams build richer customer messaging flows.

they really want those codex numbers to go up and to the right huh

A great example of a ship that was enabled by 1) our design system, telegraph and 2) agents helping us refactor out some old components that blocked the adoption without distracting the team.

@knocklabs Link here to sign up: meetup.com/nyc-elixir/eve…

Going to be showing off our new agent harness we've built at @knocklabs . Built on a virtual filesystem and powered by a bash interpreter written in Elixir. Elixir folks in NYC, come through!

@bcardarella I mean, you can technically do this today it’s just that your MCP subagent has to have its own agent loop and therefore its own provider API keys. See Sentry use-sentry tool: github.com/getsentry/sent…

MCPs should be allowed to launch subagents. There should be a single tool call for the MCP name scope along with a bit of context. Then the subagent will get the entire tool catalog the MCP supports. This would largely solve the context bloat complaints most people have.

We took all of the email API data we had and rolled it up as a handy guide for you and your agents: email.info. Powered by our Clickhouse cluster.

Read more 👉 knock.app/blog/introduci…

Agents + data functions + our messaging tools give you an incredibly powerful platform to drive product growth across all channels (including bring-your-own-components in-app messaging with guides).

Now auto classifying and approving PRs on a few of our low-risk repos via Cursor Automations.

Knock can now sit at the center of your product growth motion without needing other tools. Use the AI agent step to power enrichment, lead qualification, personalize messaging, and much more. knock.app/blog/introduci…

We added an AI step to our workflow engine to help teams build richer customer messaging flows.

@cursor_ai hugely excited about this ship! been hoping i wouldn't need to build this for our team

We're introducing Cursor Automations to build always-on agents.

We've been building a new agent mode at @knocklabs on top of the elixir just bash port. @cramforce cooked with just bash. Not all agents need a full sandbox, but most agents will benefit from operating on a filesystem with bash. Now if you're in Elixir, you can just bash.

@KINGiTAJ @cursor_ai @graphite 100% agree and feel the same

@cjbell_ @cursor_ai @graphite I really wish they would build a native mobile app. The PWA experience feels clunky at times.

I really love @cursor_ai background agents and use them regularly for oh-so-many features, explorations, and sometimes just ideating on ideas while I'm on the subway (i am a cursor mobile head). Please can we have background agents be @graphite aware so we can stack!

Our knock-cli skill makes working locally with your customer messaging an incredibly good experience. "/knock-cli Create a new email in my onboarding workflow that sends after 3 days to introduce features x, y, z." 🙌

Claude got a huge hidden update. Holy shit. Before: User prompt -> Claude -> uses tool -> Claude After: User prompt -> Claude -> writes code and logic -> that code uses a tool -> code logic can parse or format results, add conditional logic and use tool multiple times -> Claude This unlocks crazy amounts of complex function calling. For example, say you are querying a database. Previously you would do one query, then Claude would read that result and then query again if needed. Now Claude writes code to call the tool, then that code can handle the result and do different things, like query again, strip or format data, and change what it’s doing based on the tool call result, all before being sent back to Claude. The code, that Claude writes, pre plans how to react to the tool result. This compresses LLM agent loops, because the agent isn’t deciding on the fly, and it doesn’t need to keep asking the LLM to make decisions, instead the LLM pre-bakes potentially hundreds or thousands of decision paths. I would not be surprised if we see eventually 2x-100x improvements or more on agent loop & tool calling efficiency scores from this design. Subtle but absolutely huge change.

They kind of buried the lede here, but having a code execution tool and programmatic tool calling in GA now is very cool. Run code without needing to run a sandbox. Seems ideal for simple processing cases.

@stevekaliski can’t wait for the meet up to chat about minions

if you're interested in learning more i'll be talking all things agentic payments at our meetup in nyc this thursday (2/12) --> meetup.com/stripe-new-yor…

At Stripe we have a tool called "minions" -- it lets us kick off async agents built right in our dev environment to one-shot bugs, features, and more e2e. I have team, project, and personal channels dedicated just to working with minions. I like to think of it as a new type of pair programming -- "pair prompting." Read more --> stripe.dev/blog/minions-s…