Sam's Blog
An artist’s mind. An engineer’s discipline.
October 07, 2026
I am an artist. I am also a software engineer. Both are part of my DNA. Painting gave me another way to explore ideas and care about the details that make them work.
Today, I use AI to develop every piece of software I build for Hearts & Colors: our website and the internal tools that help me run our operations. This is how I work, every day. It gives a small company room to do more.
For software engineers, learning to harness AI is about more than staying relevant. It expands what we can create and gives us time for work that needs human judgement, attention and care. At Hearts & Colors, that opportunity has a purpose: more time to take care of each other.
But faster code still needs a clear intention. How do we know that what we built is what we meant?
On 7 October, I released SDD, a skill for Claude Code and Codex. SDD means spec-driven development: before writing code, describe what the software should do and how you will check it. That description—the specification—becomes the contract for building and testing the result.
For the software development community, SDD offers a way to put AI’s capabilities to work while keeping engineers responsible for the result. My skill gives the agent a repeatable process: testable requirements, human approval, implementation and evidence. You stay in charge.
The central idea is that the size of a change and the assurance a product requires are two different things. A few lines affecting payments or access rights can deserve more scrutiny than a large experiment. A quick demonstration should not carry the paperwork of a major system migration.
There are two independent dials.
Size decides the paperwork. How much is changing? That determines the specification’s structure, planning effort and approval gates, from a tiny correction to work needing separate design documents and contracts.
Assurance decides the proof. What happens if the product gets something wrong? That determines the scenarios, failure cases and verification needed. An experiment can acknowledge its limitations. Software serving customers or affecting money and safety needs stronger evidence.
The process adapts, from a one-page website to work demanding rigorous assurance.
This first release is v0. The small-change workflow works end to end across the assurance levels; inline changes have a lightweight path. Larger workflows, independent review and additional high-assurance verification are on the roadmap. Installing a skill does not make a system compliant with a regulation.
Specifications, approvals and test evidence stay with the code. You can begin in Claude Code and continue in Codex; the next session has a record to work from.
Love, care and respect guide our art and how we build this company. In software, that means respecting the people who depend on what we make. AI can help us build more; our responsibility is to make that capacity serve people.
SDD is available under the MIT licence. If you build software with AI, visit the repository, read the two-dials guide and try a small change. Issues and contributions are welcome.