Making state machines at @statelyai · prev. @Microsoft · I play piano
Orlando, FL
Joined December 2012
- Tweets 25,761
- Following 2,664
- Followers 75,749
- Likes 104,001
I'm adding Zod (Standard Schema) support to XState v6 and it looks more verbose but I don't think I had to type a full schema *once*
Just [tab] [tab]
I hope that devs stop complaining that libraries have "too much boilerplate" when they realize they can just type [tab] [tab] [tab] to generate it all
Explicit > implicit, more often than not
The combination of useEffect and useState tends to causes problems
Not because they're bad hooks (they're not; they're extremely useful), but because they're really easy to misuse
People give `useEffect` a bad rap, but it isn't usually the source of the problem. The real problem is too many `useState`s.
Devs fail to see how their data is connected and model it too discretely. As those connections appear, it is easier to synchronize than refactor the model.
And once you start down that path it becomes harder to refactor because you can't see the dependencies. You can't just look at the data declaration and be like "oh it depends on this." You have to find all the places you write to it, what those locations depend on, and what writes to those.
Even the WIP branch (without any big cleanup/optimization efforts yet) is showing a lot of code removed
I agree; if you're using mega-components that can't be decomposed, you're building with DUPLO blocks, not LEGO
Which is fine for toy examples!
I love Claude Code but I don't love getting logged out every day.
Seems to be an existing issue: github.com/anthropics/claude…
Correction: workflows are *not* DAGs!
The cycles can be implicit (e.g. retry policies or compensation logic) or they can be long-running.
Workflows are state machines at a higher level of abstraction.
Here's my 2 cents on agents !== workflows:
An agent is an autonomous entity that decides what to do next.
Agents can call workflows. Workflows can call agents.
Workflows are a DAG that describes some async task via a series of steps.
Workflows are state machines but state machines can express more than workflows (lower-level: cyclic vs. acyclic graphs).
Agent behavior can be described by a state machine (like an AI coding assistant's tool-call loop to implement a solution), including strictly following a workflow (not that "agentic", though.)
Agents are not workflows. Workflows & state machines are blueprints; agents are entities that can use those blueprints.
This is how it should be done!
AI is a mediocre developer but a solid engineer (yes, there's a difference)
David K 🎹 retweeted
we don't have to live this way. senior engineers have the duty of building guardrails and teaching the next generation of programmers the practices that lead to high quality software.
AIs may write most of the code, but you still decide what ships. we should never lose our taste for quality software.
i made this analogy a while back: everyone can cook at home but yet restaurants still exist. the floor is rising, but so is the ceiling. the best software will always be made by those who care.
Here's my 2 cents on agents !== workflows:
An agent is an autonomous entity that decides what to do next.
Agents can call workflows. Workflows can call agents.
Workflows are a DAG that describes some async task via a series of steps.
Workflows are state machines but state machines can express more than workflows (lower-level: cyclic vs. acyclic graphs).
Agent behavior can be described by a state machine (like an AI coding assistant's tool-call loop to implement a solution), including strictly following a workflow (not that "agentic", though.)
Agents are not workflows. Workflows & state machines are blueprints; agents are entities that can use those blueprints.
