--- title: The 5-Minute Agentic Engineering Maturity Score for CTOs description: 9 questions, about 5 minutes. Score your organization's Agentic Engineering maturity across Org & Team, Platform and Governance. See your level and the Wall capping your autonomy for free; leave your email for the next level, the one fix and how you compare with other teams. url: https://execuro.com/agentic-engineering-maturity-score/ updated: 2026-10-15 --- # The 5-Minute Agentic Engineering Maturity Score for CTOs 9 questions, about 5 minutes. See your level and the Wall capping your autonomy, free. Leave your email for the next level, the one fix and how you compare with other teams. [Take the score](https://execuro.com/agentic-engineering-maturity-score/) (interactive, in the browser) ## How the score works The Agentic Engineering Maturity Score places an engineering organization on a 5-level ladder, from No Adoption (level 0) to Autonomous Engineering (level 4). It scores three dimensions with three questions each. 1. Each answer is tied to one level, 0 to 4. 2. A dimension's score is the average of its three answers; its level is that score rounded down. 3. Your overall level is your lowest dimension: the weakest dimension caps how much work you can hand to agents. 4. Your Wall is what blocks that lowest dimension from reaching the next level. ### The three dimensions - **Org / Team:** How your people and work are structured - **Platform:** The foundation you build and ship on - **Governance:** Risk, trust, and accountability ### The five levels #### Level 0: No Adoption Ad hoc use, throwaway output **What gets delegated to agents:** Effectively none. AI is autocomplete; a human controls every line and nothing is delegated as a whole unit of work. #### Level 1: Experimentation First “aha” moments, trials **What gets delegated to agents:** Minimal. Prompt-and-draft works, but a human rewrites the result and reviews every line; no agent owns an end-to-end task. #### Level 2: AE Anchoring Anchored mindset, tools, rulesets **What gets delegated to agents:** Supervised execution (HILT-steer). A human specs a unit of work, the agent executes it, and the human watches for and corrects drift. #### Level 3: AE Scale Practice scales across teams **What gets delegated to agents:** Gated autonomy (HILT-approve). Agents run spec → build → verify; humans approve at defined gates and trust verification over reading diffs. #### Level 4: Autonomous Engineering Self-improving at scale **What gets delegated to agents:** Governed autonomy. Agents self-initiate from the backlog within guardrails; humans supervise via telemetry + policy, not by reading outputs. ### The Walls What holds an organization at each level, per dimension. Your result names the one that applies to you. #### Org / Team - **Level 0 → 1:** The team hasn’t had its first “aha” moment, and PMs don’t really understand what the engineers are talking about. AI is still just fancy autocomplete, and the org runs exactly as before: specialists in fixed lanes, work moving by handover, the business queuing behind IT’s feature factory. - **Level 1 → 2:** You’ve seen the magic — two or three AI-fluent people are visibly faster — but it stops with them. Roles are still rigidly siloed, only a few have AI fluency and boundaries hold, and work still moves by handover. So the wins stay as individual and local sparks, not yet the team’s over SDLC. - **Level 2 → 3:** Mixed signals: some teams/roles genuinely speed up, others quietly stall on the old process, and it all depends who’s on the team. Wins live in roles instead of a standard outcomes across involved roles/teams. - **Level 3 → 4:** You have real builders who own requirement→release; but builders still live inside IT, the business can’t self-serve, and engineers aren’t embedded where the work is. You’ve built a faster factory, not yet part of the business. #### Platform - **Level 0 → 1:** No harness and no platform — raw IDE autocomplete, each dev wires their own environment, every deploy is a manual one-off. Nothing you set up makes the next project any faster. - **Level 1 → 2:** Solo wins don’t transfer. 2–3 early adopters per team have built their own clever setups, but there’s no shared Agentic Harness and nothing org-wide to build or deploy on — so every good idea stays solo and the next person starts comes with a diff. mental model. - **Level 2 → 3:** First Agentic Harness is anchored, not mature, not product, and IT is still the provisioning/deploy bottleneck, environments are hand-made. You’ve documented golden paths, but need an environment or a deploy? You still file a ticket and wait. Nothing is self-service yet. - **Level 3 → 4:** Self-service works — internal apps go repo-to-prod without a ticket — but harness and platform aren’t yet one integrated, agent-operated surface, and deploy/data/identity aren’t governed by policy-as-code. Humans still drive every step; agents can’t act on their own. #### Governance - **Level 0 → 1:** Ungoverned — output ships unchecked, no provenance or accountability; risk is invisible. You’d only learn something broke after it already has. - **Level 1 → 2:** Your safety net is a human reading every output line by line — and verify-everything doesn’t scale. Review is manual and strains under throughput, and nothing is gated on entry, so the more AI produces the more your best people drown in review. - **Level 2 → 3:** You’ve got real discipline now — nothing ships until it clears acceptance criteria. But review load scales 1:1 with throughput — humans are the bottleneck because everything is still gated on entry. Your people have become the ceiling on how fast you ship. - **Level 3 → 4:** You’ve broken the volume problem — agent-verified work ships and humans review only flagged exceptions. But trust still depends on flagged-exception review, with no telemetry/policy supervision or self-improving loop. So when something slips, you fix that one case and it can happen again.