The thesis
“The question is badly framed. You do not choose between the core and the agents; you choose, domain by domain, which rules must become legible before an agent is allowed to act on them.”
— David Alejano, Chief Strategy Officer at Thinkia
What decides is not the system but the legibility of its rules. An agent that reads from an opaque core is a contained risk; an agent that writes to it is comprehension debt at machine speed.
What you will learn
- Why replacing the core first and putting agents on top fail for opposite reasons, and why the usual mistake is funding them in the wrong order
- The four questions that place each domain: where the rule lives, whether the agent reads or writes, what happens if it gets it wrong, and how often the rule changes
- The four-quadrant framework (agents on top, contract first, decode in parallel, core first) and what gets funded in each
- The governed-contract layer between the core and the agents: per-action permissions, per-invocation traceability and cost control
- Five ways to get it wrong, from the heroic rewrite to autonomy that grows faster than the ability to audit it
- The 2027 budget in three lines funded in order (decode, contract, execute) and the five indicators for the committee
Who this is for
CIOs, CTOs, enterprise architects and executive committees who must split the 2027 budget between modernising the core and deploying agents, with a criterion that survives an audit.