I'm working on something similar as a side project. I'm finding frustration with a lack of repeatability in my LLM flows. 90% of my code is AI written, but most of my guidance to LLMs is not particularly specific. It's "make sure you've read this file", "how does that match against existing patterns", "what's the performance like".
I've ended up building my workflow engine directly in Python, despite YAML being the default choice for LLMs.
I found that YAML had some drawbacks:
* LLMs don't have an inherent understanding of YAML conventions. They tend to be overly verbose. Python code solved this because "good" code is generally as short as you need.
* YAML isn't really composable. Yes, you can technically compose it, but you'll be fighting the LLM the entire time. Python solved this because the LLM knows how to decouple code.
* I want _some_ things to be programatic still. Having Python solves that
* Pretty much any programming language would do. Python just feels like the default for LLM-centric code.
I've always been of the view that for a workflow language, you should use a proper, turing-complete functional language which gives you all the usual flexiblity for transformations on intermediate data, while also supporting things like automatic parallelisation of things like external, compute-intensive tasks.
Do you implement a DAG within your system to act as a kind of well-defined backbone for analysis and execution, or do you dispense with (explicit) DAGs entirely?
Here's a different kind of workflow engine with a proper DSL. It turns out config management is the same problem as workflow engines, if you use my modern definition of config management.
I was expecting to see some verbose LLM output, but actually the code has a distinctly hand-crafted feel. Nice to see! I'm not sure if "production ready" is a safe claim 7 commits in to a project ;)
Was going to ask the same thing. The orchestration space already has some very well established frameworks like Airflow and Dagster. Would be curious to see the pros and cons.
I think the future of replacements to well established frameworks written in Python/etc.. are zero dependency binaries (from Rust or Go) that require so little configuration and tuning and they "just work".
Type-safe code. Workflows are not configuration! If I wanted YAML hell I could stick to Github Actions.
But that's only the start. There are a lot of other things I would expect of a new workflow orchestrator in 2026 so if you are not comparing yourself to the competition you probably don't know what you're getting yourself into.
Yeah, that makes sense. I looked at a few workflow orchestrators and I'm building something that I will release soon, but my thinking is that the "workflow engine" should be an abstraction that takes the input and executes the steps. "What" you use to define that workflow is probably the SDK layer though, but I can certainly see the value in using type safe code to define as opposed to a YAML file.
I'm mainly focusing on the portability aspect of it (e.g. use TS/Python/etc. to define the workflow/steps or just simple a simple YAML file).
Sort of. My thinking is that the input to define the workflow should be anything you prefer to use (TS, Go, YAML, etc.) and the orchestrator's job is to model that and execute the job, given your deployment model.
DAG Workflow Engine
A production-ready DAG (Directed Acyclic Graph) workflow engine driven by a YAML DSL. Validates, executes, and visualizes workflows with support for parallel execution, retries, conditional branching, batch iteration, and pluggable actions.
My version of a similar tool; but written in Go with a compile time guarantee.
https://github.com/swetjen/daggo
I'm working on something similar as a side project. I'm finding frustration with a lack of repeatability in my LLM flows. 90% of my code is AI written, but most of my guidance to LLMs is not particularly specific. It's "make sure you've read this file", "how does that match against existing patterns", "what's the performance like".
I've ended up building my workflow engine directly in Python, despite YAML being the default choice for LLMs.
I found that YAML had some drawbacks:
* LLMs don't have an inherent understanding of YAML conventions. They tend to be overly verbose. Python code solved this because "good" code is generally as short as you need.
* YAML isn't really composable. Yes, you can technically compose it, but you'll be fighting the LLM the entire time. Python solved this because the LLM knows how to decouple code.
* I want _some_ things to be programatic still. Having Python solves that
* Pretty much any programming language would do. Python just feels like the default for LLM-centric code.
I've always been of the view that for a workflow language, you should use a proper, turing-complete functional language which gives you all the usual flexiblity for transformations on intermediate data, while also supporting things like automatic parallelisation of things like external, compute-intensive tasks.
I recommend checking out https://github.com/peterkelly/rex and also my PhD thesis on the topic https://www.pmkelly.net/publications/thesis.pdf.
The gap in flexiblity between DAG-only and a full language designed for the task is a significant one.
Do you implement a DAG within your system to act as a kind of well-defined backbone for analysis and execution, or do you dispense with (explicit) DAGs entirely?
Here's a different kind of workflow engine with a proper DSL. It turns out config management is the same problem as workflow engines, if you use my modern definition of config management.
https://github.com/purpleidea/mgmt/
How does this compare to Temporal? That seems to be the current baseline for application-oriented workflow engines.
I was expecting to see some verbose LLM output, but actually the code has a distinctly hand-crafted feel. Nice to see! I'm not sure if "production ready" is a safe claim 7 commits in to a project ;)
https://github.com/vivekg13186/Daisy-DAG/blob/main/backend/s...
I've seen LLMs include that exact "production-ready" claim on code they generate. But of course it gets that from its training data.
What makes it production ready? What's the code coverage on your tests? There are only seven commits in this repo as of this comment.
how it compares to airflow?
Was going to ask the same thing. The orchestration space already has some very well established frameworks like Airflow and Dagster. Would be curious to see the pros and cons.
I think the future of replacements to well established frameworks written in Python/etc.. are zero dependency binaries (from Rust or Go) that require so little configuration and tuning and they "just work".
That being said, that's not this project.
I don't see any references to existing orchestrators, which are way more complete, so I presume you did this as an exercise?
Just seeing YAML used for workflows in this age makes me automatically nope out.
Curious, what format would you prefer to use to represent a workflow instead of YAML?
Type-safe code. Workflows are not configuration! If I wanted YAML hell I could stick to Github Actions.
But that's only the start. There are a lot of other things I would expect of a new workflow orchestrator in 2026 so if you are not comparing yourself to the competition you probably don't know what you're getting yourself into.
Yeah, that makes sense. I looked at a few workflow orchestrators and I'm building something that I will release soon, but my thinking is that the "workflow engine" should be an abstraction that takes the input and executes the steps. "What" you use to define that workflow is probably the SDK layer though, but I can certainly see the value in using type safe code to define as opposed to a YAML file.
I'm mainly focusing on the portability aspect of it (e.g. use TS/Python/etc. to define the workflow/steps or just simple a simple YAML file).
Are you planning to map those varied definitions onto varied orchestrators?
Sort of. My thinking is that the input to define the workflow should be anything you prefer to use (TS, Go, YAML, etc.) and the orchestrator's job is to model that and execute the job, given your deployment model.
DAG Workflow Engine A production-ready DAG (Directed Acyclic Graph) workflow engine driven by a YAML DSL. Validates, executes, and visualizes workflows with support for parallel execution, retries, conditional branching, batch iteration, and pluggable actions.