What AI-ready actually means for an EPD team

There's a test we run at the start of every AI readiness assessment, and it has nothing to do with AI. Imagine a strong senior engineer joins your team on Monday. Using only what's written down, the docs, the tickets, the architecture decisions, the codebase, could she ship something meaningful by Friday without interrupting anyone?

In most organisations the honest answer is no, and that answer predicts AI outcomes almost perfectly. An agent is the most literal-minded new hire you will ever onboard. Everything your documentation assumes a human would just ask about, an agent either can't do or does wrong.

Readiness is a property of your system, not your stack

After running readiness assessments across scale-ups and enterprises this year, we'd reduce AI readiness to five measurable properties of the delivery organisation:

1. Context accessibility. Decisions, constraints, and reasoning exist somewhere machine-readable: architecture decision records, written product rationale, current docs. Teams whose real knowledge lives in Slack history have nothing to point a model at, and no amount of licensing fixes that.
2. System access. Agents act through APIs. If your core systems are reachable only through a UI built in 2011, agents can observe your business but not participate in it. Integration is the most cited blocker for teams putting agents into production, usually discovered after the agents were bought.
3. Eval capability. You can define what a correct output looks like: explicit definitions of done, testable acceptance criteria, ground truth for the judgments you want to delegate. Without it, you can't tell whether the system works beyond anecdote.
4. Decision latency. Agents compress execution time, so slow human approval loops become the dominant term immediately. A team that takes two weeks to approve a change will watch agents idle for two weeks, just faster.
5. Ownership. Every automated action has a named, accountable owner for the permissions behind it and the rollback when it misfires. Organisations that can't name an owner for a workflow today won't produce one when it's automated.

What's deliberately not on the list

Model choice, vendor, prompt libraries, a Chief AI Officer. Those are procurement details. The five properties above are organisational, they take months to build, and they're exactly what the Friday test measures: a written-down, API-accessible, evaluable, fast-deciding, clearly-owned delivery system is the environment where any new senior, human or otherwise, becomes productive in a week.

Why readiness work pays even without agents

Every one of the five properties makes a human team faster on its own. Written decisions cut onboarding time. API access speeds every integration. Eval discipline catches regressions. Fast approvals and clear ownership compound across everything. The agent simply collects the dividend at a higher rate, which makes readiness the rare AI investment with no downside scenario.

How to grade yourself before Q4

Run the Friday test honestly, then score the five properties for your own organisation. The gaps you find are your actual AI roadmap, and most of them cost discipline rather than money:

• If context lives in heads, start recording decisions with reasoning now
• If core systems lack APIs, that integration backlog is your real AI budget
• If "correct" is undefined, eval work starts with product, not engineering

For the graded version with a prioritised fix list, a three-week Framed readiness assessment produces exactly that, and it's the cheapest insurance against funding agents your organisation can't yet use. If you'd rather start with a conversation, talk to us; more on how we run delivery is in our insights.