Overview
What this research program is and why it exists. The frame the rest of the work hangs on.
research / devplane
DevPlane research
Two products in one project — a local cockpit for multi-tool software development (kanban + actor-handoff protocol + coordination-event log + MCP server) and a portfolio-wide shared engineering brain (pattern library, architecture maps, session handoffs, cross-project assignment registry, capability-architecture doctrine). Research is an empirical program on coordination cost in heterogeneous AI tool ecosystems — using DevPlane's continuous production telemetry as the apparatus, not the subject. Lead study: a pre-registered field test of risk compensation in human-AI coordination.
Why this matters
The portable claim — what this research lets you understand outside the surface domain.
The productivity claims being made for AI coding tools are largely grounded in agent-side measurements — lines produced, tasks completed, time-to-PR. If the Ironies of Automation (Bainbridge 1983) are operative — operator vigilance falling as agent reliability rises — those measurements systematically overstate net effect. The DevPlane research program tests that prediction with continuous production telemetry on a real operator running real agents on a real, multi-month codebase. The methodology generalizes: any team running heterogeneous tools through a coordination layer (multi-tool ops dashboards, hospital handoff systems, distributed scientific instruments) shares the same shape of problem.
Read first
The general-audience explainer is the entry point. Everything below is the drill-down.
DevPlane research — overview
General audienceCoordination cost in heterogeneous AI tool ecosystems. Outside-reader brief on what the program studies, why now, and what it contributes.
Read →Drill-down — full research surface
Seven-slot baseline. Forthcoming slots shown openly.
Methodology
How the work is done — instruments, protocols, the standards each report inherits.
Reports
The actual research findings — phased results, research-question briefs, applied analyses.
Coordination cost in heterogeneous AI tool ecosystems — program
The full three-arm research program: Agent↔Agent, Human↔AI, and Interface (the lead arm). C1 — risk compensation in human-AI coordination — is the lead study. Theoretical lineage across cockpit HCI, CSCW, empirical SE, behavioral decision-making, and stigmergic-coordination literatures.
Feature brainstorm convergence (2026-05-02)
Five parallel cluster lenses (coordination, working alliance, trust calibration, governance, identity) synthesized 143 raw features from the AI-research corpus into a Now/Next/Later/Killed slate. Documents the convergence map (which primitives appeared in 3+ clusters) and the decision rule that filed 14 cards into the shipping wave.
Phenomenology delta — attention-as-duration & skill formation (2026-05-02)
Cross-references the post-2010 phenomenology-of-attention literature (Stiegler, Citton, Hayles) and phenomenology-of-skill (Dreyfus, Merleau-Ponty, Heersmink) against the 14-card synthesis slate. Surfaces what the existing measurement set covers (verification load, withdrawal disruption) and what it misses (sustained reasoning duration, situation-disclosing experience, transparency-vs-incorporation). Source for DP-88 (sustained-reasoning window) and DP-91 (phenomenology probe).
Cross-application AI-surface mapping (2026-05-02)
Inventory of AI surfaces across vela and meta-factory, mapped to the 14-card synthesis slate. Identifies which devplane primitives port directly to other portfolio apps (origin marker, multi-axis reviewer, sycophancy circuit-breaker) and which require app-specific adaptation. Source document for the cross-repo dispatch wave (DP-93 through DP-96).
Audience tiers
The same headline research surfaced four ways: peer-review, engineering, general audience, product.
General-audience explainer
General audiencePeer-review framing
Peer-review framingEngineering critique
Engineering critiqueAdversarial review of the DevPlane research instrument: the two-phase actor handoff, the coordination-event log schema, multi-tool sync drift across claude-code/cursor/codex, and the completion-block protocol. Names specific failure modes (fire-and-forget write drops, build-agent-skips-review shortcut, agent_started step-function pre/post DP-102, polling-bound merge detection) and a ranked fix list before the corpus carries weight.
Product implications
Product implicationsWhat the research program tells a product builder thinking about coordination tooling for human-AI software development. What to build into DevPlane next, what to build into adjacent tools, what the broader product space gets wrong — traceable to architectural decisions and pre-registered predictions, with honest pre-data caveats where applicable.
Bibliography
Field positioning — formal references and literature maps grounding the research threads.
Preregistrations & protocols
Studies and intervention protocols filed before execution.
Pipeline
What is running, what is queued, what is forthcoming.