About this site
Who writes it
This site is written and maintained by Technoidentity, a Durable Product Engineering firm. We build software for companies where an unexplainable decision is not an embarrassment but a regulatory event. Much of what we build now is agents, on systems that were running long before the agent arrived and will run long after.
The definition here is the book’s. It is versioned. The landscape page is dated and corrected. The score page stores nothing. Cases on this site are composites drawn from real engagements, with identifying details changed.
There is no analytics on this site and no third-party scripts. Nothing you enter on the score page leaves your browser.
The assessment
One agent workflow, assessed against the five properties. You get a written report: where the workflow stands on each property with evidence, what blocks it from production, what would have to change, and what that would cost. Two to three weeks. You keep the report whether or not you work with us afterwards.
Corrections
If something on this site is wrong, tell us which line and why. Corrections are made and dated, and the landscape page lists them.
The first correction on this site was ours. The book says the pilot-failure figure is consistent across IDC, MIT and Digital Applied. The 14% comes from one survey. The other two measure different things and point the same direction.
22 September 2026
/about
Corrected the book's claim that the pilot-failure figure is consistent across three analysts. See sources.
Sources
Every figure on this site carries a footnote, or it is cut. What each source measures is stated, because the three below measure different things.
- 1
Digital Applied, The AI Agent Scaling Gap: From Pilot to Production, March 2026. Survey of 650 enterprise technology leaders, VP and above, across manufacturing, financial services, healthcare, retail and professional services, fielded February and March 2026.
Measures The share of enterprises with agent pilots and the share at production scale, and the gaps respondents named most often. The unit is enterprises, not pilots.
The five gaps are multi-select responses. They are what respondents named, not shares of a total.
- 2
Gartner, press release, 25 June 2025. Over 40% of agentic AI projects will be cancelled by the end of 2027, due to escalating costs, unclear business value or inadequate risk controls.
Measures Forecast project cancellation, not pilot-to-production.
- 3
MIT NANDA, The GenAI Divide: State of AI in Business 2025. 95% of generative AI pilots show no measurable P&L impact.
Measures Generative AI broadly, not agents.
We can find no canonical publisher page for this report. It circulates as a version 0.1 PDF re-hosted by third parties. It is listed here because the book refers to it, and it is treated as directional only. No figure on this site rests on it.
Contact
The firm’s work on agents that write to systems of record is at technoidentity.com/durable-agents. The definition and the test stay here.