We been helping mid-market companies for the past 1.5 years and finally ready to share the internal platform publicly.
Yansu (严肃) is a AI coding platform that use spec + TDD to build complex software projects. It is more like a SOP than coding agent. We focus on understanding requirements and checking outcomes against those requirements while iterating the code based on the tests.
Yansu tries to learn as much tribal knowledge as possible. These are things you don’t write down in google doc or Notion. Yansu absorbs these knowledge by continuously talking to users and distilling learning from them.
It's as if a spec + TDD platform had a baby with character.ai.
Why care about requirements? B/c 80% of any software development is understanding requirements and what exactly we want to build. We also focus on outcomes, the only thing that matters. We deliver satisfying outcomes by simulating scenarios and generating tests based on those scenarios. Our agents take that tedious testing part of the code away from others.
We prioritize accuracy over latency/cost, using a mixture of agents (not limited to CC, codex, and etc) to get the job done. We then run through continuous-testing-generating pipelines until all things pass.
What does Yansu mean? “Serious” in Chinese. Just like my favorite artist, Rene Magritte, painting in his kitchen in a suit. I want to give my coding respect and care.
Our goal is to level people up from IC to tech leads, to work on the high leverage work of planning, validating, and educating.
We made a launch video to celebrate human work that builds on all of the creative minds before us. Shoutout to CinemaSings for making this happen.
Referring to LLM-usage as "teaching" is both kinda-insulting to actual teaching--a task not entirely new to senior engineers--and also a rather grand exaggeration of LLMs ability to be-taught.
Twiddling with prompts is not "teaching", it's guess-n-check whack-a-mole until you get something barely good enough to ship, at least temporarily... until it Disregards All Previous Instructions and barks like a chicken or the training data shifts enough that you need to throw on a new set of arbitrary influential phrases.
The lack of easy analogies for controlling an LLM isn't really because it's an amazingly good control scheme, but because it's so weirdly unreliably awkward that it's something humans don't even try to make in our other machines and systems. It'd be like designing an industrial stamping machine so that its controls were buttons on a Twister™ sheet: It'd be quite novel, but not in a very good way.
You have a wrong assumption of what context and teaching means with twiddling with prompts.
The job of human is providing teaching via feedback. Your manager does the same thing to you too. And you can distill those feedback into learnable experience for your llm to be better next time.
As we are building a collaboration platform between you and your AI coworkers. We wrote down our experience on this new relationship, and how it reflect in our product, where are the relationship now and how they are evolving over time.
I think the target personas is different. While they might have the same capabilities, but the job-to-be-done is different.
openllmetry is focus on engineers, who wants to use this as more of a piping solution and it sits on top of opentelemtry. While opentelemetry is a popular solution. It is just applying a solution to a new problem.
OpenLayers to me is thinking from the ML/AI problems from ground up and while serving the data scientists and probably prompt engineers.
The tea from a region in Fujian has a great flavor that is referred to as "duck-feces fragrance". When people ask about the soil, the farmers say they use duck feces to deter others from stealing their soil.
Turn that exploratory work to product would be the challenges. It is hard to balance the two