Oystro . ADLC protocol for AI coding agents

Bring any agent. Keep your standards.
Open-source, agentic development lifecycle protocol for AI coding agents: human gates + project memory. Free.

[Oystro . ADLC protocol for AI coding agents @oystro]

Building Oystro, an open-source agentic development lifecycle protocol for AI coding agents. Human gates before code, evidence before release. Works with any agent. https://t.co/OLwuXBN8tw

[George Koshy @georkos]

@Grady_Booch @steipete A software engineer who thinks his job is to bring value to the stakeholders will adopt workflows/protocols that help LLM write better code faster.

[George Koshy @georkos]

When you start on a new project, an agent will impress you. On a brownfield project with 15,000 lines of code, the same agent will guess. It will contradict itself and break what matters. The agent hasn't gotten worse. The missing context is now obvious.

[George Koshy @georkos]

People are averse to touching codebases that are left to be fixed. Do not focus on perfect code. Before you start coding, you should decide what needs to be built and what is out of scope. An agent can dodge that question. You cannot.

[George Koshy @georkos]

@andrewchen Founder ships the first version with agents. Then hands it over to a real engineer who inherits a codebase nobody has read and has to make the product survive.

[George Koshy @georkos]

"We will document it later" is a promise the agent won't keep. It doesn't remember making it.

[George Koshy @georkos]

An agent session ends and takes everything with it. The constraints you found, the approaches you rejected, the reason for the shape of the code. You have two choices. Be the memory yourself, or leave it in the repo where the next session reads it. One of those scales. The other exhausts you.

[George Koshy @georkos]

Read the plan before the code. Read the evidence before the release. Two habits induce most of the reliability.

[Oystro . ADLC protocol for AI coding agents @oystro]

Every agent session starts from zero. Oystro keeps the decisions and the evidence in the repo, so the next session can continue. Open source and works with any agent.

[George Koshy @georkos]

https://t.co/HHziaEFIfn

[George Koshy @georkos]

Multi-week projects don't fail loudly. They drift. Nobody notices until a feature ships that nobody can explain.

[George Koshy @georkos]

Most teams keep waiting for a better model. Teams that ship reliably already have a good enough model. What separates them is everything around the model. What context it gets, where a human steps in, and what gets recorded. The process is the difference, not the IQ.

[George Koshy @georkos]

The agent never gets tired of being told the same thing. You do. That's the actual bottleneck.

[George Koshy @georkos]

Agents are great at producing code and terrible at telling you what they left out. Ask what's deferred, every time.

[George Koshy @georkos]

The longer the project, the more the agent's confidence outruns the evidence. Review the evidence, not the confidence.

[George Koshy @georkos]

"One task at a time, one commit per task" is not ceremony. It's the cheapest audit trail you will ever get. When something breaks three weeks later, the commit that caused it is one screen, not one day of `git blame`.

[Z47 @z47_vc]

"If you cannot measure it, you cannot improve it"
In this episode of #MatrixMoments we cover the essentials of tracking startup metrics & why we think it's a vital tool in decision making, hear what @tarun_davda & @VikramV23 have to say: https://t.co/YsrLbIOWo8 @rbalaraman

[BNI® Global Network @BNI_official_pg]

#MondayQuote from Dr. Ivan Misner

[Australian Space Agency @AusSpaceAgency]

The Australian Space Agency has signed a Statement of Strategic Intent with Adelaide-based start up @MyriotaGlobal. https://t.co/jiUrxKqYHf