Scope & delegate
Make the project, resource, environment, and proposed plan explicit before execution begins.
- Evidence
- Codex project/worktree; Replit plan; Retool resource choice
External product evidence · mechanism audit · 2026-09-03
How leading builders make consequential work visible.
A first-party visual study of how developer tools support creation, inspection, testing, deployment, governance, and recovery. This dated sample supports comparison of visible mechanisms; it does not measure product reliability, adoption, or user outcomes.
60Exact product states
13Products
6Reusable mechanism lenses
Primary finding
The strongest systems keep intent, artifact, evidence, commitment, and repair in one inspectable loop.
Intent→Manipulable artifact↔Evidence→Human decision→Action→History + repair
Six mechanism lenses
Compare what the interface lets people understand and control.
These lenses organize visible mechanisms across products. They are a comparison framework, not a universal workflow, interface, or architecture.
Create & steer
Keep the initiating intent connected to a visible artifact that can be corrected directly.
- Evidence
- AI Studio prompt/preview; Stitch streaming canvas; n8n node editor
Inspect & compare
Join output with the trace, diff, input, query, or reference result needed to judge it.
- Evidence
- Codex diff; LangSmith trace/evals; Temporal event history
Approve & govern
Separate recommendation from authorized action and surface policy conflicts at the decision point.
- Evidence
- Palantir staged writes; Copilot publication block; Retool audit logs
Release & hand off
End creation in a familiar, inspectable artifact: pull request, deployment, publish, export, or package.
- Evidence
- v0 PR/preview; AI Studio deploy; Stitch export; Workato package
Recover & resume
Provide named checkpoints, scoped rollback, rerun, and durable execution history.
- Evidence
- Replit rollback; Workato rerun; Temporal history; AI Studio versions
Synthesis for exploration
A legible path from orientation to repair.
The evidence supports a comparison model, not a universal screen sequence. It helps identify missing states, unclear authority, weak correction, or unsupported leaps from inference to action.
01 · OrientScope and contextowner · project · resource · environment · known state
02 · ExpressIntent and constraintsrequest · plan · success criteria · exclusions
03 · ManipulatePrimary representationartifact · code · canvas · graph · object · workflow
04 · InspectEvidence and alternativespreview · diff · trace · query · test · comparison
05 · CommitReview and authoritypolicy · approval · staged change · accountable decision
06 · OperateRelease and handoffpublish · deploy · PR · export · package · downstream owner
07 · RepairHistory and recoverycheckpoint · rollback · rerun · changed facts · audit trail
The representation is the trust surface.
Conversation becomes more reviewable when it stays coupled to the artifact people are actually responsible for understanding and correcting.
Evidence belongs near the decision.
Diffs, traces, inputs, outputs, and tests reduce switching cost only when they remain connected to the exact change under review.
Governance works best before action.
Staging, policy checks, and visible authority boundaries are more useful at commitment time than as a downstream audit surprise.
Recovery is a designed user journey.
A checkpoint name is not enough; people need to know which state domains will change and whether repeated side effects are safe.
Handoffs should produce familiar objects.
Pull requests, deployments, packages, and owned catalog entities make generated work addressable by existing teams and processes.
Similarity is not validation.
Repeated patterns establish category precedent. They do not establish customer value, technical feasibility, compliance, or the best representation for another context.
Product ledger
Thirteen products, organized by observable state.
The sample spans agentic development, prompt-to-app builders, visual orchestration, governed operational logic, evaluation, catalogs, and durable workflows.
Evidence atlas
Every claim stays attached to a visible state.
Use the controls to narrow by product or search the observed state, action, artifact, or mechanism. Each card includes what is visible, the available action, the artifact, source links, and evidence limits.
Filter by product
Method boundary
All selected images come from first-party public pages, documentation, demos, or authorized public training material. Website chrome was removed by using the original embedded product asset whenever available; video frames were cropped deterministically. No evidence image was generated, retouched, inpainted, or otherwise altered. Observation date: September 3, 2026. Some source media predates that review; the Temporal image is from a 2023 workshop. Product versions and account entitlements were not independently tested. Source descriptions and visible controls do not establish reliability, adoption, accessibility, or user outcomes.