Agile after the framework wars

Cadence beats branding. What still works when nobody wants another SAFe rollout is not a secret and never was.

1. A two-week heartbeat with a real demo.
2. A backlog someone can say no to.
3. A definition of done that includes the operator, not just the ticket.
4. Retros that change a working agreement, not a mural.

That is the whole religion. The rest is merch.

ANZ has spent years on agile at a scale that filled coaching benches across Australia and New Zealand. Shopify spent years as the poster child for product cadence and then for deleting meetings that did not earn their keep. GovTech Singapore ships digital services inside government constraints with a professionalism that makes a lot of private-sector "agile transformations" look like costume. Atlassian wrote the Team Playbook so squads could steal rituals without buying a new religion. Four different animals. One useful skeleton.

How we got a merchandise aisle

Agile was a set of values and a few practices that made software honest. Then it became a transformation market. The market needed names, roles, certifications, and wall art. SAFe gave large organisations a picture that fitted an org chart. Spotify's old squad/tribe diagram gave everyone else a picture that fitted a keynote. Consultancies sold both, sometimes in the same quarter.

ANZ-style programmes got value when squads could demo working slices to a product owner who could say no. They got theatre when the picture of the operating model became the work. Shopify-style product orgs got value when cadence and ownership were real. They got theatre when other companies copied the diagram and not the decision rights. GovTech Singapore got value because the service had to work for a citizen and pass a standard. Atlassian got value because playbooks are defaults a team can edit, not a compliance badge.

If your agile programme can name the framework faster than it can name last fortnight's demo, you bought merch.

A backlog someone can say no to is the whole religion. The rest is merch.

Cadence, defined without a vendor

A heartbeat. Same length, long enough to finish a thin slice, short enough that a miss is cheap. Two weeks is a default, not a commandment. Some GovTech and platform teams use longer. Some product teams use shorter. The crime is a heartbeat that changes every time a steering pack gets nervous.

A demo. Working software or a working service slice. Not a slide about the software. ANZ squads that still matter demo. Atlassian teams demo. Shopify teams ship. If the demo is a narrative of blockers, you had a status meeting.

A backlog with a no. One person (or a tight pair) can kill or defer. If everything is priority one, you have a wish list. SAFe PI planning can produce a honest no. It can also produce a room that cannot say no in public. The artefact is not the point. The no is the point.

A definition of done that includes the operator. The ticket is not done when the branch merged. It is done when the store, the banker, the citizen-facing officer, or the SRE can run it. GovTech Singapore already thinks this way because the user is outside the building. Banks forget. Retailers forget. Law firms never started.

A retro that edits a working agreement. One rule changes, or you do not meet. "We will not accept work without an owner." "We will demo even if we are embarrassed." "We will not book status on top of the heartbeat." Write it down. The Atlassian Team Playbook is full of stealable versions. Use one. Edit it.

What to retire

The scaling picture as the programme. You may need a coordination forum. You do not need a poster of trains to start saying no.

Story-point theatre. If the conversation is about the number and not the slice, you are estimating to soothe a dashboard.

Agile coaches who never attend the demo. Coaching is a residue in the working agreement, or it is a cost centre.

Ceremonies with no artefact. Stand-up that does not surface a stuck slice. Retro that does not edit a rule. Planning that does not produce a no.

Copying Shopify's meeting purge without copying Shopify's writing habit. Deleting the room without a written decision path is how work goes to chat and dies there. Steal both or steal neither.

ANZ, Shopify, GovTech, Atlassian: what to steal (not what to copy)

From ANZ-scale banking. Respect for risk in the heartbeat. The useful pattern is a risk or ops counterpart who sees the demo, not a stage-gate that resets cadence every quarter. Steal the counterpart. Do not steal a three-year operating-model programme unless you also steal the demo discipline.

From Shopify. Cadence, ownership, and a willingness to delete rooms that do not produce decisions. Steal the deletion test: "What decision does this meeting make?" If the answer is none, write instead. Do not steal a cargo-cult of "we are like a startup" inside a licensed bank.

From GovTech Singapore. Standards and reuse as design inputs. A service that must work for a citizen will not be saved by a tribe diagram. Steal the standard questions. Steal the habit of showing the thing.

From the Atlassian Team Playbook. Rituals as defaults a team can edit. Health monitors, working agreements, demo habits. Steal a page. Run it for two heartbeats. Keep or kill. That is agile. The merch store is optional.

A working week after the wars

Monday: planning that ends with a no and a slice list the operator would recognise.

Mid-heartbeat: one stuck-work forum, 25 minutes, only items that have been stuck for more than a day.

Friday (or the last day): demo to a sceptic. Retro that edits one rule. Residue note the same day.

Everything else is opt-in. If ANZ, a Shopify-style product org, GovTech Singapore, or an Atlassian squad needs more rooms, they can add them. They should have to justify them to the heartbeat, not to a framework badge.

A reset for a tired programme

If you are a CIO, a CDIO, or a Head of Agility who has inherited a merch wall:

1. Pick three squads. Not thirty.
2. For two heartbeats, require a real demo and a written no.
3. Add the operator to the definition of done.
4. Make the retro produce one edited rule, published.
5. Kill one standing meeting that does not make a decision.
6. Report residue (slices shipped, rules changed, meetings killed), not maturity scores.

If that reset works, expand. If it does not, you did not have an agile problem. You had an ownership problem. Frameworks cannot fix that. Cadence can expose it.

Collective Campus runs Agile T&E as cadence practice, not as a religion shop. We will teach a playbook page. We will not sell you a train.

Definition of done, written for an operator

Software teams love "code complete." Operators need "I can run this on a Tuesday."

Write done as:

• The slice works in the environment the operator uses (store device, banker desktop, citizen service, SRE runbook).
• The operator was shown it in the demo, or a recorded two-minute walkthrough exists.
• The failure mode is named (what we do when it breaks).
• The metric we expect to move is named.
• Support or the floor knows who to call.

ANZ-scale banking needs this because a feature that surprises the branch is an incident. GovTech Singapore needs this because a citizen cannot raise a Jira. Shopify-scale product needs this because a merchant will tweet. Atlassian playbooks already push demo and visibility. Steal the operator line even if your playbook page does not have it yet.

If adding the operator slows the heartbeat, good. You were shipping unfinished work and calling it velocity.

Coordination without a train

Large organisations still need to line up dependencies. You do not need a scaled framework badge to do it.

• A written dependency board updated in the heartbeat, not in a quarterly theatre.
• A 25-minute forum for stuck cross-team work only.
• A named integrator on the rare programme that actually is a programme (a core cutover, a regulatory date).
• A rule: no new coordination room without deleting one.

ANZ has enough history with scaling pictures to know which forums were useful (the ones that unblocked a slice) and which were costume. Shopify-style orgs delete first and add later. GovTech Singapore coordinates through standards and reuse more than through tribe diagrams. Atlassian coordinates through visible work and playbooks. Steal those. Leave the merch.

What to tell the board instead of a maturity score

Boards get sold "agile maturity" because it is a number that can only go up. Useless.

Report:

• Percent of squads that demoed working slices this heartbeat
• Number of explicit nos in planning
• Operator-included dones
• Working-agreement edits
• Recurring rooms killed
• Cycle time on one journey the board already cares about

If those numbers are ugly, you have a real programme. If a maturity score is green and those numbers are empty, you have merch. ANZ, Shopify, GovTech SG, and Atlassian teams can all produce the first list. Make your vendor do the same, or change vendors.

What to keep after SAFe

A lot of ANZ-scale, NAB-scale, and Lloyds-scale programmes bought a scaled framework because the org chart needed a picture. Some of the picture was useful. Most of the merch was not. After the wars, keep the parts that make a fortnight honest. Retire the parts that make a poster.

Keep a heartbeat that does not move when a steering pack gets nervous. Two weeks is a default. Some GovTech Singapore and platform teams use longer. The crime is a cadence that resets every time a committee wants certainty. SAFe PI planning can produce a honest no. Keep the no. You do not need the train poster to keep the no.

Keep a visible dependency board updated in the heartbeat. Large organisations still have dependencies. You do not need a quarterly theatre to list them. A written board plus a 25-minute stuck-work forum is enough for most ANZ and Telstra-scale delivery. Add a named integrator only when you actually have a core cutover or a regulatory date.

Keep a product owner (or a tight pair) who can kill. If PI planning produced a room that cannot say no in public, you kept the wrong part. Shopify-style product orgs and Atlassian playbooks already treat the no as the religion. Steal that. Leave the tribe diagram.

Keep risk or ops in the demo, not in a stage-gate that pauses cadence. ANZ-scale banking got value when a counterpart saw the slice. It got theatre when the operating-model picture became the work. Keep the counterpart.

Keep a working agreement the team can edit. SAFe and every other scaling picture will try to standardise rituals into compliance. Atlassian's Team Playbook is the steal: defaults a team can change. If a rule cannot be edited in a retro, it is merch.

Retire maturity scores, story-point dashboards as the programme, coaches who never attend the demo, and any coordination room that does not unblock a slice. If you cannot name last fortnight's demo faster than you can name the framework, you kept the wrong religion.

GovTech Singapore already coordinates through standards and reuse more than through tribe diagrams. Shopify-style orgs delete first and add later. Steal those habits. You can do that while still running a planning increment if the increment produces nos and slices. The badge is optional.

A practical keep-or-kill note for the CDIO:

We will keep the heartbeat, the demo, the no, the operator in done, and a written dependency board. We will retire the maturity score, the merch wall, and one coordination room that did not unblock a slice last quarter. If delivery slows, we will restore a room that was actually a build. If it does not, we were paying for a picture.

Demo hygiene

A demo is not a narrative of blockers. A demo is a working slice a sceptic would recognise. ANZ squads that still matter demo. Atlassian teams demo. Shopify teams ship. GovTech Singapore teams show the thing because a citizen cannot raise a Jira. Hygiene is how you stop the demo becoming a status meeting with better lighting.

Show the environment the operator uses. Store device, banker desktop, citizen service, SRE runbook, care desktop at Telstra. A localhost happy path is a rehearsal. Say it is a rehearsal or do not call it a demo.

Time-box to the slice, not the saga. Fifteen minutes of showing. Five minutes of "what is not done." Five minutes of the no: what we will not start next heartbeat. If you need 45 minutes of context, you did not write a pre-read.

Invite a sceptic. Risk, ops, a floor lead, a service owner. If only friends attend, you scheduled applause. Capital One-style design orgs already know this. Banks forget.

Name the failure mode. What we do when it breaks. Who the floor calls. If that sentence is missing, the slice is not done. See operator-included done.

Record a two-minute walkthrough if the operator could not sit. Do not replace the live demo with a deck. Replace an absence with a walkthrough, then get the operator into the next one.

Circulate residue the same day. What shipped, what failed in front of people, what we will not do next. No sequel by default.

If the demo is embarrassing, good. You learned in public while the miss was cheap. If the team refuses to demo until it is polished, they are running a stage-gate inside an agile costume. Call it that.

The no in planning

Planning without a no is a wish-list reading. SAFe rooms, Spotify-model copies, and "SAFE-lite" programmes all fail the same way: everything is priority one, so nothing is. The artefact is not the point. The no is the point.

Write the no into the working agreement.

1. One person or a tight pair owns the no. Product owner plus a constraint counterpart if you are licensed (credit, conduct, clinical, comms). A room of twelve cannot kill.
2. Start planning from capacity, not from the incoming list. If the list does not fit, the no happens in the room, out loud, written.
3. Say the no in customer or operator words. "We will not start the portal this fortnight. We will finish the status SMS." ANZ and NAB customers do not experience your PI. They experience the seam you deferred.
4. Publish the no next to the yes. A backlog someone can read. Atlassian-style visibility. If the no lives only in a private corridor, you will get shadow work.
5. Do not take "just a spike" as a way around the no. Spikes have a kill date and an owner. Infinitev-style and Capital One-style rooms already know cheap tests. Unowned spikes are how frameworks rot.

Monday planning that ends with a slice list the operator would recognise and a written no is agile. Monday planning that ends with a commitment wall and a smile is merch.

A practical planning close:

Read the nos first. Then read the slices. If we cannot name three things we will not start, we are not finished. Residue in 15 minutes: yes list, no list, owner of each, demo date. Then stand up.