The spiritual successorto DORA.
HX builds on the frameworks that came before it rather than arguing with them. This page lays out what DORA, SPACE, and DX Core 4 established, what changed when agents arrived, and why reading AI-era work takes a new instrument.
Three frameworks taught the industry to see its own work
HX keeps their discipline and their findings. Everything on this site is built on them — cite them, keep running them, and hold HX to their standard.
Two things happened that no prior framework was designed for
Between 2024 and 2026, the day-to-day of engineering changed shape: much of the work now happens as sessions with AI agents. That broke two assumptions every earlier instrument was built on.
A new primary source appeared
For as long as we've measured software work, there were two kinds of evidence: artifacts — commits, pull requests, deployments, tickets — which record what happened; and surveys, which record how it felt, weeks later, on average. Sessions are a third kind: a verbatim record of how the work actually went. What was asked, what was tried, what failed, where it stalled — the reasoning, not just the result.
Frameworks built before this record existed measure around it: their unit of analysis is the artifact or the quarter. HX's unit of analysis is the session.
In an agentic organization, artifacts record what happened and sessions record why. An instrument that only reads artifacts misses more of the work every quarter.
The loop can finally close
Measurement used to live beside the work — an analytics silo fed by exports, read in reviews, acted on by memo. That was a tooling limitation, not a law of nature, and the agent era removes it: when work is conducted on a platform that can also read the work, a finding can return to the next session — as coaching, a better prompt, a fixed process — while the context is still warm.
HX therefore requires what earlier frameworks could only recommend: measure inside the system of work, and close the loop where the work happens.
Findings that have to travel between systems rarely change anything. Findings that return to the next session do.
What HX keeps, what HX changes
- Outcomes over activity. DORA's discipline, kept as a house rule: no activity leaderboards, ever.
- The multidimensional view. SPACE's five dimensions echo in CARE's four questions; no single number carries the answer.
- Felt experience as data. The survey tradition, kept honest: every check-in earns a visible outcome.
- Deployable definitions. Core 4's lesson: a framework spreads when its numbers are publicly specified.
- The unit of analysis. From artifacts and quarters to the working session — a human and their agents, together.
- The cadence. From quarterly recall to a continuous record, with felt check-ins on the person's terms.
- The consent model. From a policy appendix to constitutive architecture: covenant, aggregates, access log.
- The home. From an analytics silo to the system of work itself, so insight returns to the next session.
| Framework | Year | Unit of analysis | Primary evidence | Cadence | Consent |
|---|---|---|---|---|---|
| DORA | 2014 | The delivery pipeline | Deploy & incident telemetry | Continuous telemetry, annual reading | Not needed — it measures machines |
| SPACE | 2021 | The developer & the team | Telemetry + periodic surveys | Quarterly, typically | Guidance, not architecture |
| DX Core 4 | 2024 | The engineering organization | Mixed telemetry + self-report | Quarterly benchmark | Policy-level |
| HX | 2026 | The working session — a human and their agents | The work's own record + felt signals | Continuous record; check-ins on the person's terms | Constitutive: charter, k≥5 aggregates, access log |
Successor, with care
None of this makes DORA wrong, SPACE obsolete, or Core 4 misguided. Each read the best evidence available in its decade. But when most of the work happens in conversation with agents, the best evidence of how work is going is the conversation itself — and none of the earlier instruments can read it, because it didn't exist when they were built.
- ✕That delivery metrics are obsolete. Keep the four keys — they read the pipeline well. HX reads what they can't.
- ✕That measurement alone improves anything. HX requires the loop: findings must return to the place work happens, or the reading is theater.
- ✕That the covenant is optional. An implementation without the trust architecture isn't a partial HX. It's monitoring, whatever the pricing page says.