From source code to shared intent
The old abstraction was code over machine language. The next abstraction is intent over code.
That does not make engineering disappear. It moves engineering upward. The hard work becomes stating the real problem, carrying context without distortion, choosing constraints, testing behavior, and deciding what is worthy of release.
An artifact is not a chat response or a mockup. It is a working object: a page, tool, system, document, model, or interface that can be opened, examined, used, and improved. The conversation is scaffolding. The artifact is the proof.
Not prompting. Engineering.
It optimizes the next response.
It manages intent, construction, verification, and revision as one system.
Seven principles
Intent is the first material.
Before choosing tools, state what must become true for the person using the artifact. A vague request can begin the work; it cannot finish the specification.
Context is part of the design.
Constraints, history, audience, environment, and taste are not overhead. They determine whether the right thing gets built.
The artifact outranks the conversation.
Fluent explanation does not compensate for broken behavior. What opens, runs, reads, calculates, or communicates is what counts.
Verification is a creative act.
Testing reveals what the artifact actually is. Inspect the output, exercise the interactions, challenge assumptions, and revise from evidence.
Human judgment remains accountable.
Synthetic intelligence can generate options and execute at speed. People still own purpose, consequences, acceptance, and release.
Legibility is a feature.
A good artifact makes its operation and limits understandable. Hidden cleverness is worth less than visible reliability.
Iteration compounds.
Preserve what was learned. Each artifact should make the next one easier to specify, safer to build, and faster to verify.
The working method
Artifact engineering is a loop, not a single generation.
FrameName the change
Define the user, the situation, and the outcome that matters.
GroundGather reality
Bring in the source material, constraints, interfaces, and facts the work depends on.
ShapeChoose the form
Decide what the artifact should be, what it should omit, and how someone will use it.
BuildMake a complete object
Produce the smallest finished artifact that can honestly satisfy the intent.
ProveTest the claim
Inspect the rendered result, exercise its behavior, and compare it with the original outcome.
CarryKeep the learning
Record the decisions and patterns that should survive into the next iteration.
A research agenda
A discipline needs more than a name. It needs questions that can produce methods, evidence, and tools.
How can ambiguous human goals become precise without flattening what people mean?
How should a system preserve context while keeping it current, bounded, and under human control?
What evidence makes a generated artifact trustworthy enough for its consequences?
How can artifacts explain failure and support correction without requiring their makers to start over?
How should credit and responsibility work when human judgment and synthetic production are inseparable?
What becomes possible when the ability to make software is no longer limited by fluency in syntax?
The artifact engineer’s pledge
I will begin with human intent, build toward a real outcome, test what I make, reveal what I cannot prove, and take responsibility for what I release. I will use synthetic intelligence to enlarge human agency—not to disguise the absence of judgment. The artifact will be my evidence.