Writing
This is the publishing layer for the project: shorter notes, longer essays, and references that build the broader argument around post-chat AI interfaces. Pattern demos and side-by-side examples live on the pattern pages, not here.
Long-form
4 essays
Oct 08, 2026
What has to be true before you let a system act on your behalf. Four conditions read along time, and the seven patterns chat never covered.
Aug 25, 2026
You designed for someone watching. Now nobody is. Which patterns hold up, and the evidence grade behind that claim.
May 09, 2026
Twelve major platforms converged on the same chat box. Where the conversational interface stops working, and what replaces it.
May 09, 2026
Directability, Observability, Predictability, Explainability. The four lenses every pattern here is graded on.
Published
Each piece pairs an argument with a single diagram. They live first on LinkedIn and then settle here, linked back into the patterns they argue for.
6 essays
The DOPE framework for trustworthy AI autonomy
Four overlapping properties — Directability, Observability, Predictability, Explainability — define whether an AI system is safe to delegate to. Their intersection is trustworthy autonomy.
“No audit without transparency. No enterprise AI without audit.”
Connects to patterns
From conversation design to delegation design
Conversational UIs optimize for the next reply. Delegation UIs optimize for handoff, monitoring, and intervention — a different surface entirely.
“The design question shifts from helpfulness to trustworthiness.”
Connects to patterns
Not binary. Not one-size-fits-all.
Copilot, delegation, autonomous — three points on a continuum users move along as trust accumulates. Actual control and perceived control both matter; they are not the same axis.
“Control isn’t binary. It shifts with task, experience, and risk.”
Connects to patterns
When everything looks the same
The same model run by different people produces aesthetically convergent results. The cure is not better prompting — it is deliberate deviation from the statistical mean.
“More individual creativity, less collective variety.”
Connects to patterns
The designer’s role in the age of generative AI
Generative AI compresses the cost of producing variants. The designer’s job shifts from drawing every option to choosing among many — and that selection itself is the work.
“AI spits things out, but without my direction, the result has no soul.”
Connects to patterns
Design as connective tissue
AI dissolves the sequential handoff between Product, Design, and Engineering. Where they overlap is judgment — and design becomes the discipline that holds it.
“The one function that asks “should we?” instead of “can we?””
The Core Argument
The problem is not that chat is bad. The problem is that it was designed for one thing — question and answer — and is now being asked to carry something much heavier.
When you work with an AI on anything iterative — writing, research, planning, code — you need to do three things chat cannot do well: inspect what the system is actually doing, steer the process mid-stream, and build on previous decisions without re-reading a transcript.
Every prompt is a fresh instruction. Every constraint is managed by you. Every mistake requires another message. The system is capable. The interface is not keeping up.
What Comes After
Visible State
Plans, memory, confidence levels, and reasoning traces should be surfaces — not buried in the thread or hidden behind the output.
Direct Intervention
Users need to approve, pause, rewind, and constrain. A text box is not an intervention point. It is a new instruction that may or may not be understood.
Persistent Context
The constraints and decisions from message 1 should be as visible in message 47 as they were when they were written.
Structured Control
Sliders, toggles, panels, timelines, and approval gates are not regressions to old UX. They are the right tool when the work requires precision.
The Newsletter
The Beyond Chat newsletter — each new AI UX pattern and the thinking behind it, sent as it ships. One considered email, roughly weekly.