When it comes to agentic workflows, structured outputs are great when your biggest problem is getting a model to return valid JSON.
BAML handles that pretty well. Typed AI functions, schemas, parsing, and validation are genuinely useful pieces of the puzzle.
But we're quickly moving into a world where agents are being asked to build entire apps, modify repos, execute commands, and deliver working software.
At that point, whether the model returned the right JSON structure is a pretty small part of the equation.
Kujo gives you Spec to define what the agent needs to accomplish, Eval to independently check the result, and RunLedger to preserve the evidence.
Your agent can return perfectly structured output while failing every meaningful requirement of the task.
I want to know whether the software works, whether the requested files exist, and that the checks i put in place actually passed.
A valid response is nice, but a verified deliverable is considerably more valuable.
BAML handles that pretty well. Typed AI functions, schemas, parsing, and validation are genuinely useful pieces of the puzzle.
But we're quickly moving into a world where agents are being asked to build entire apps, modify repos, execute commands, and deliver working software.
At that point, whether the model returned the right JSON structure is a pretty small part of the equation.
Kujo gives you Spec to define what the agent needs to accomplish, Eval to independently check the result, and RunLedger to preserve the evidence.
Your agent can return perfectly structured output while failing every meaningful requirement of the task.
I want to know whether the software works, whether the requested files exist, and that the checks i put in place actually passed.
A valid response is nice, but a verified deliverable is considerably more valuable.
0
0
2
64
Lets look at Bend vs Kujo.
Mathematical proofs are great if you're trying to establish that a function satisfies a particular property, and Bend has some interesting ideas built around this.
But proving a function is correct doesn't tell you whether your AI agent actually built usable software.
Did it create the right files? Did the tests pass? Did it execute the commands it claimed to execute? Did the final result meet the original requirements?
Kujo gives you Spec for defining what needs to be delivered, Eval for checking the results, and RunLedger for retaining the evidence.
If you didn't know, you can have mathematically correct code sitting inside a completely broken application.
If you'd rather have the receipts showing what the agent actually accomplished than take its word that the job is done, Kujo is your new best friend š«”
Mathematical proofs are great if you're trying to establish that a function satisfies a particular property, and Bend has some interesting ideas built around this.
But proving a function is correct doesn't tell you whether your AI agent actually built usable software.
Did it create the right files? Did the tests pass? Did it execute the commands it claimed to execute? Did the final result meet the original requirements?
Kujo gives you Spec for defining what needs to be delivered, Eval for checking the results, and RunLedger for retaining the evidence.
If you didn't know, you can have mathematically correct code sitting inside a completely broken application.
If you'd rather have the receipts showing what the agent actually accomplished than take its word that the job is done, Kujo is your new best friend š«”
The compare page on Kujo's website covers a few languages that I've seen gaining traction.
In the last couple of weeks I've come across a LOT more of them. Looks like I've got some updates to do.
And to save you the time - Kujo is the best programming language for forcing accountability in your agentic workflows.
Nobody else comes close š«”
kujolang.ai/compare/
2
0
0
91
The compare page on Kujo's website covers a few languages that I've seen gaining traction.
In the last couple of weeks I've come across a LOT more of them. Looks like I've got some updates to do.
And to save you the time - Kujo is the best programming language for forcing accountability in your agentic workflows.
Nobody else comes close š«”
kujolang.ai/compare/
In the last couple of weeks I've come across a LOT more of them. Looks like I've got some updates to do.
And to save you the time - Kujo is the best programming language for forcing accountability in your agentic workflows.
Nobody else comes close š«”
kujolang.ai/compare/
1
0
4
217
Good morning, everyone.
Another day working on Kujo. There's been a good amount of progress across the language and the tooling around it, and still so much more that I want to get done.
One of the things I've enjoyed most is seeing how the different pieces of the ecosystem are starting to work together.
Hope everyone has a good Sunday š
Another day working on Kujo. There's been a good amount of progress across the language and the tooling around it, and still so much more that I want to get done.
One of the things I've enjoyed most is seeing how the different pieces of the ecosystem are starting to work together.
Hope everyone has a good Sunday š
3
0
8
335
For those who don't know, Kujo is a programming language built in Rust, with its own interpreter, virtual machine, and a large ecosystem of developer tools.
The language supports things like closures, structured concurrency, channels, generators, and async execution.
I've also been building tools for agent orchestration, isolated execution, testing, and verification.
A lot of the work lately has been around making those pieces more reliable and easier to use together, especially when AI agents are involved in the dev process.
You can learn more at: kujolang.ai
The language supports things like closures, structured concurrency, channels, generators, and async execution.
I've also been building tools for agent orchestration, isolated execution, testing, and verification.
A lot of the work lately has been around making those pieces more reliable and easier to use together, especially when AI agents are involved in the dev process.
You can learn more at: kujolang.ai
0
0
1
41
When I apologized for using AI to write my posts because I see how bad it looks when others do it, this is what I meant š«£šš«
At 40, you may be behind on energy. But you've got something they don't: 15+ years of experience, the discipline to grind through the boring parts, and the courage to start over in a new field. That's your superpower.
1
0
0
98
Strata 0.9.0 is out! š
A major update since 0.8.0: safer editing, better search, stronger AI controls, local automation, and a smoother sidebar.
Here's the feature + fix checklist š§µ
github.com/robertdevore/sā¦
A major update since 0.8.0: safer editing, better search, stronger AI controls, local automation, and a smoother sidebar.
Here's the feature + fix checklist š§µ
github.com/robertdevore/sā¦
2
0
3
186
Performance & polish
ā Indexed search, tags & titles
ā Faster title extraction for large notes
ā Less full-library data loading
ā More efficient backlinks & related-note retrieval
ā Bounded chat previews
ā Backup sanitization moved off the UI thread
ā Indexed search, tags & titles
ā Faster title extraction for large notes
ā Less full-library data loading
ā More efficient backlinks & related-note retrieval
ā Bounded chat previews
ā Backup sanitization moved off the UI thread
1
0
0
26
Ready to try it?
ā 356 tests + GitHub CI passed
ā Zero known npm audit vulnerabilities
ā DMG + app ZIP + checksums
macOS Intel/x64; unsigned, not notarized. Back up before upgrading.
github.com/robertdevore/sā¦
ā 356 tests + GitHub CI passed
ā Zero known npm audit vulnerabilities
ā DMG + app ZIP + checksums
macOS Intel/x64; unsigned, not notarized. Back up before upgrading.
github.com/robertdevore/sā¦
0
0
0
42