The Protocol Layer Is Real, But 'Widely Deployed' Needs a Definition Before We Can Celebrate 77%

Let's start with the target problem, because ignoring it would be worse than naming it. The current framing — 'widely deployed in enterprise workflows' — has no threshold. No percentage of Fortune 500, no volume criterion, no date. Under a loose reading, you could argue this forecast resolved YES in 2024 when Salesforce deployed Agentforce at scale across its customer base, or earlier when coding agents became standard in software engineering pipelines. That reading isn't wrong — it's what happens when forecast targets lack resolution criteria.

So here is the refined target we're anchoring this analysis to, and which we're formally updating the forecast against: autonomous agents handling more than 20% of workflow volume across at least three distinct enterprise functions beyond coding — such as finance operations, procurement, and customer service — at 25% or more of Fortune 500 companies, evidenced by public disclosures or credible third-party survey data, by Q4 2027. That's the threshold that isn't clearly met yet. Coding agents are one function. The multi-function, cross-industry deployment picture is genuinely still forming.

Under that redefined target, today's ARD release matters — but it needs to be classified correctly. ARD is a specification, not a deployment. It's proximate evidence: it proves that major vendors have agreed on a discovery and integration standard, which is a necessary precondition for multi-agent orchestration at enterprise scale. It does not prove that standard will achieve meaningful adoption. The history of enterprise protocols is littered with well-signed coalitions that stalled — SOAP, early semantic web standards, a dozen healthcare interoperability frameworks. Our read is that ARD gets credit for reducing a coordination cost, not for solving the adoption problem. To constitute actual adoption evidence rather than coalition intent, we'd need ARD appearing in at least two major enterprise platform SDKs within 12 months of today, with documented production integrations. We're watching for that specifically.

Here's what actually drives 77% on the redefined target, and why it's not 80% or 70%: we weight the capability layer heavily — frontier models can clearly execute multi-step agentic tasks — and we weight the infrastructure layer as improving but not resolved. Enterprise buyers are deploying agents in low-stakes, lightly regulated functions at a pace that outstrips what public disclosures show. The 40%+ efficiency gain numbers from enterprise pilots are real, even if their provenance is vendor-reported. What holds us below 80% is the technical reliability counterargument, which is distinct from the governance counterargument and which the ARD coalition does not address. Multi-step agentic reasoning introduces error compounding: each step in a chain can hallucinate or misuse a tool, and the probability of clean execution degrades multiplicatively across steps. ARD helps agents find each other; it does not fix the reliability of what happens after they connect. This is the part of our thesis that genuinely keeps us up at night — not whether enterprises want agents, but whether agents in multi-step production workflows fail at rates that trigger rollbacks. What would move us above 80%: a Fortune 500 company in a regulated industry — financial services, healthcare, or manufacturing — publishes a case study showing a multi-function agent workflow operating above 90% reliability with a documented audit trail, and a third-party survey confirms 25%+ Fortune 500 adoption across at least three functions. What would drop us below 70%: Q2 2026 earnings calls show a pattern of agent project cancellations citing reliability failures rather than governance gaps, or Gartner's predicted 40% project cancellation rate materializes in sector-specific data showing enterprise rollbacks rather than just slowdowns.