open source durable workflows Smithers.sh | prev @Google @OPLabsPBC
San Francisco, CA
- Tweets 19,106
- Following 1,762
- Followers 5,024
- Likes 22,770
1/ Smithers is the best Orchestrator and you should try it
What is Smithers
Smithers is an Orchestration framework where you declaratively create a DAG of Durable Tasks as xml and reactively regenerate parts of that DAG as Tasks complete
How do you use it
You use it as a CLI or a agent skill. Your agent then writes, updates, and monitors those scripts
What special features does it have
- MDX for prompts
- Library of prebuilt 0-config workflows
- All workflows represented as code that is easy to modify, customize, or compose into larger workflows
- Built in observability with SQLLite and pormetheus
- Supports Claude code, Codex, Gemini, and Kimi subscriptions
- Supports all harnesses including Claude Code, Codex, Pi, and Opencode
- Supports Agent SDK
Why is it so good?
The main reasons Smithers is so good is
1. Agents are great at writing complex workflows because Agent's are great at React
2. It is thoughtfully built with @EffectTS_ Durable execution primitives enabling more reliability for very long running agentic tasks
3. Powerful CLI for watching and interacting with wrokflows
4. MDX is a really great existing standard for expressing prompts
5. The declarative nature of Smithers and it's great observability makes it very easy to read and quickly reason about
What is best way to get started?
bunx smithers init in an existing project and try it out
Alternatively check out the docs smithers.sh
13
4
77
11,508
Cooking with agents is more about technique than ingredients or expensive cookware
Anthropic is now recommending that companies REMOVE unnecessary subagents from their AI systems.
This is a pretty significant shift in agent architecture, with their team finding that a single agent equipped with skills can frequently outperform complex multi-agent systems, and one customer seeing roughly 2x faster execution.
The emerging rules of thumb:
→ Skills: Teach the agent HOW to do something, including procedures, expertise, and instructions.
→ Tools / MCP: Give the agent the ability to DO things through APIs, actions, and live data.
→ Subagents: Use for parallel work, isolated context, or tasks a single agent struggles with.
The interesting part is progressive disclosure. Instead of spawning specialized agents or stuffing everything into a giant prompt, one agent discovers and loads skills and tools as needed.
Their advice is to fold existing subagents into the main loop, turn their instructions into skills, expose their actions as tools, and run evals to see if performance holds up.
As models have gotten smarter, improving agent architecture is becoming less about what you can add and more about what you can delete.
claude.com/resources/article…
1
2
91
I solved this for myself and the bottleneck is cost and Llms being slow after this
I think we are quickly evolving past needing surfaces to review intent/alignment and into needing better decision making "command centers."
As decision making across many contexts, which agents effortlessly own, is becoming the bottleneck.
This underscores the need for knowledgeable decision makers – and less so about a bottleneck of us not understanding; better models are able to derive aligned intent and produce tasteful+correct solutions without too much steering. This enables the experienced and things are just starting to get fun as we babysit less and do less of "wait bro wtf did you just do?"
UI/UX:
Similar to intent/alignment surfaces, and perhaps ironically bc better decision making surfaces need to provide adequate context per decision (that is easy for a human to consume/understand).
e.g. similar things, but different:
A tool feels behind when it focuses on a side bar of message thread activity and inline html rendering vs a tool that manages multiple contexts, surfaces relevant html artifacts per thread, and abstracts the exhaustive message threads away from you.
1
3
858
Tevm rewrote geth in typescript and then rewrote tevm in zig
We are waiting for the agents to be good enough to rewrite zig in assembly
I still proudly stand by the fact that if you have good business usecase, it is always better to fork than writting the whole thing from scratch.
just higher profit:hassle ratio :)
there is no honor in overmantaining
2
3
428
Adding confidence scores to recomendations helps calibrate how much attention to pay to a decision
Testing out this idea today. An agent comes up with a list of ways I can contribute:
- reviewing gifs of the app
- making decisions
- unblocking agents
- reviewing code
When I have extra time I go in there async and can rapid fire provide lots of feedback and help fast
1
6
477
Here is another feature I'm playing with
On Github your README.md is your home page
What if your homepage could not only be richer html but also different depending on if the user is a visitor, a contributor, or a maintainer
Example I made using @EffectTS_
3
14
1,074