Engineering Beyond Agile: Speed With a Spine
Introducing the ISEE Framework
The operating model for everything this series has been pointing at
It’s been seven articles. Seven things that start to fall apart when execution outruns us. Thank you to the awesome leaders trusting me with their concerns, ideas, experiments. Listening to my evolving ideas. Truly, you get me to open my work laptop every single day.
A buffer disappears. A decision arrives too late. Context fragments across a team that can no longer wait for alignment. A trade-off hardens before anyone had the conversation. A platform despite high expectations is still a service catalogue. A deployment slides past a boundary the organisation never encoded. A “reasonable output” reaches production while nobody noticed the choice was made.
Each piece started somewhere different. They ended in roughly the same place. The problem is not the speed, but rather the absence of a structure that speed can run inside.
I’ve been circling this shape since the first article without quite being able to put my finger on it. Part of the reason is that frameworks announced too early tend to become slogans before they become useful. Part of the reason is that I wasn’t sure it was a framework at all — it might have just been a pattern I kept noticing.
After seven articles, and many conversations, it has earned a name.
Intent · Structure · Execution · Evidence
I’ve been calling it ISEE. Four layers. Two directions of flow. One operating model for what an Agentile organisation actually is.
What the series actually built
If you haven’t read the previous articles, I would recommend doing so. It helps to lay the pieces next to each other before drawing lines between them.
The first article named the condition. Agentile — agentic execution and human-agentic collaboration replaces the Agile process. The system that ships while you sleep. I wrote about the pressure this puts on ceremonies, hiring, and the shape of the team itself.
The second asked what happens to the product manager when AI removes the buffer. The PM does not disappear. The PM moves upstream, toward describing intent and outcomes.
The third followed the movement upstream to its conclusion. When execution becomes cheap, late decisions become expensive. Every ambiguous requirement, every unstated constraint, now travels further and faster before anyone notices.
The fourth introduced the concept of the cell. The minimum unit that can ship with full context. Not a stage in a pipeline. A locally accountable node in a system that no longer behaves like a flow.
The fifth made the trade-offs structural. Security, cost, capacity — no longer review topics, because the review happens after the decision has already been made. Codifying them is not bureaucracy. It is turning a judgement call into a constraint while it is still cheap to change.
The sixth named the organisation that makes this possible. Platform engineering stops being tooling and becomes the place where human intent is encoded into the system that executes.
The seventh asked the hardest question of the season. If humans can no longer be in every loop, what must remain in human hands? Accountability. Direction. The never-delegate list.
Delegate work. Not responsibility.
That was the closing line of Article seven. It is also the load-bearing claim underneath everything that follows.
The shape that emerged
ISEE is not a process. There are no ceremonies, no timeboxes, no rituals to adopt. It is a description of where things live in a fast organisation — and a claim that those things live at different speeds, on purpose.
Figure 1 — The four layers of ISEE, with the cadence at which each one moves.
Intent is the never-delegate list. Strategic direction. The things the organisation will not compromise on. It moves slowly, because it should. Quarters and years, not sprints.
Structure is how Intent becomes enforceable. Codified trade-offs. Golden-path defaults. Policy-as-code. Specifications and architectural decisions kept alive rather than archived. Structure is the spine — slow to change, instant to enforce. This is where article five lives. This is where article six lives. This is where every decision that used to live in a meeting now lives instead.
Execution is where the work gets done. Agents write code. Humans direct it. Cells ship continuously. Execution is fast. It is meant to be.
Evidence is what the running system says about itself. Behavioural telemetry. Audit by construction. The layer that closes the loop back to Intent, because reality carries more information than the meeting that preceded it.
Four layers. Different cadences. Different owners. Getting the layers right is not an optimisation. It is the difference between a system that learns and a system that merely accelerates.
The central claim is simple: They do not move at the same speed. The asymmetry is the design.
The asymmetry of speed
Agile assumed speed was a uniform property of the team. Make the team faster and the work goes faster. Sprint planning became more precise. Standups became tighter. The whole model was optimised to reduce friction — on the assumption that friction was the bottleneck.
In an Agentile organisation, execution is no longer the bottleneck. The bottleneck is the slow layers: Intent and Structure. The places where judgement lives. The places that cannot and should not move at machine speed.
Figure 2 — Humans sit on the slow layers. Agents run the fast ones. The spine is the translation between them.
Humans sit on the slow layers. Agents run the fast ones. The spine is the translation layer between them — human intent, rendered once, enforced many times per second.
Speed ≠ velocity. Fast execution inside a weak structure is just faster failure.
Here is what that looks like at 2am on a Tuesday. An agent opens a pull request that passes every test, every policy gate, every cost check. It merges itself. It deploys itself. By 02:14 it is serving traffic. By 09:00 someone notices the new behaviour — a subtle shift in how the system treats a specific class of customer — and asks the reasonable question: who signed off on this.
Nobody did. Nobody needed to. The spine said yes, and the spine was telling the truth about a decision nobody consciously made.
That is Intent leaking into Evidence without ever passing through Structure.
The system did exactly what it was told. The telling is what needed more humans in it.
The loop that closes it
A model with four layers and no feedback is just a diagram.
Figure 3 — ISEE as a loop. Evidence is what turns four layers into one system.
Evidence is what makes ISEE a loop rather than a waterfall. It is the layer that challenges Intent — not because the system is critical of the humans running it, but because reality always has more information than the meeting that preceded it.
Here is what that looks like when it works. A leadership team commits to a direction in January. By March, telemetry shows their customers are routing around it. Not complaining. Not churning. Just quietly using a path nobody designed. The spine faithfully enforced the January decision for ten weeks. Evidence is now, politely and insistently, telling the leadership team that the January call is no longer the right one.
If the team changes direction in April, the loop closed. If they don’t — because the roadmap is committed, because the narrative has hardened, because nobody wants to be the one who says the January call was wrong — the spine begins to enforce a fiction.
A spine enforcing a fiction is the most expensive kind of alignment you can buy.
The test of a healthy ISEE organisation is not whether it ships fast: It is whether it learns.
What Changes on Monday Morning
If you have followed this series and want to know what to do with it, here is where to start.
Figure 4 — Four questions. One for each layer of the operating model.
Name your Intent layer. What is on your never-delegate list? If you cannot name three things your organisation will not compromise on — under speed, under pressure, under commercial urgency — Intent is implicit. Implicit intent gets encoded into the spine by whoever is closest to the keyboard.
Test your Structure layer. Is your platform a service catalogue, or is it a spine? A service catalogue provides options. A spine encodes decisions. Ask the question: if a new cell spins up today and plugs into your platform, what constraints does it inherit automatically? If the answer is “not many,” you have a service catalogue.
Hunt for ghost decisions in your Execution layer. Every week, your system makes thousands of choices nobody remembers making — rollback thresholds, dependency bumps, feature-flag rollouts, environment selections. Pick one. Ask who owns it, what the constraint was that produced it, and whether that constraint is still true. If the answer to any of the three is “I don’t know,” you have a ghost decision. Ghost decisions are where Intent quietly moves from Structure into Execution without anyone approving the move.
Read your Evidence. Not the dashboard. The signal. What is the system telling you that contradicts something you believed a quarter ago? If Evidence never challenges Intent, one of two things is true: either the organisation is unusually well-aligned, or nobody is actually reading it.
Closing thoughts
ISEE is not finished. The stress points are real. Multiple spines that encode contradictory trade-offs. Intent that dilutes as it cascades to cells. Evidence that becomes a firehose too loud to read. Roles — the engineering manager most of all — under pressure from every direction.
These are not theoretical risks. They are the next set of problems this blog series will face.
If you want to be an early reviewer of the ISEE Framework, let me know. All I ask is your feedback, either written or in a discussion.
Season 2 of the blog series tackles these problems and puts real people inside them. Not frameworks and layers. Roles. The security professional who used to be a gatekeeper and now has to author the guardrail. The engineer whose craft is shifting from authorship to orchestration. The platform team that used to serve developers and now serves agents too. The architect whose decisions run at machine speed. The engineering manager whose composite role is being pulled apart by every layer of the model. The executive who must lead from Intent without losing the thread to Evidence.
Every article in Season 2 will ask the same question this season has been asking all along.
What does this actually mean for your Monday morning?
The answer has not changed.
Delegate work. Not responsibility. The spine does the rest.
Next in the series — Season 2
Season 2 opens in the room nobody wants to be in. The post-incident review where the agent did everything right, the policy held, the rollback worked — and the customer still lost data. The question is no longer who is at fault. It is which line of Structure was missing, and who was supposed to have written it.
Meet that person first.