Benjamim Castell

“We accept the risk.” Does that include accepting a later delivery if it happens? Or is the team still expected to hit the original date? I'd want to know that before treating it as a decision.

Naming an engineer the “risk owner” won't change another team's priorities. Following up is useful. But if both teams need the same capacity, who gets them to agree on which work comes first?

The dashboard shows where consequences landed. The decision log shows where authority sat. Record decision, authority, consequence. Otherwise a decision made outside the team can return as a team failure. Full argument: benjamimcastell.substack.com/p/agile-didn...

Your team has a dashboard. Leadership has no decision log. Velocity, carryover, and missed dates stay visible. Scope freezes, late approvals, and unresolved dependencies can vanish. That is how accountability becomes one-sided.

A useful archive entry records two things: why the item left the active backlog, and what observable evidence would bring it back. That keeps the history recoverable without pretending the commitment is still current. What brought your last restored item back?

Archive ≠ delete. Keep the request, context, and reason it mattered. Remove only its active status—the unsupported promise that the team still intends to build it. New evidence can bring it back.

A simple test for an old “priority”: name the observable change that would move it up. No condition? Keep the history. Remove the active promise. Bring it back when the evidence changes.

“Priority” is not a storage label. An old item earns active status only when someone can name the current problem and what would make it rise. No answer? Preserve the record. Archive the promise.

Do not run a backlog purge. Run a decision test. Keep active: current problem + current goal + owner + evidence + selection condition. Archive: searchable record + decision date + return condition. Then measure what comes back with new evidence.

Take the oldest item in your backlog. Ask: • Current problem? • Current goal? • Decision owner? • Recent evidence? • What would make it rise? No clear answers? Preserve the record in a searchable archive. Stop letting it impersonate an active choice.

Deleting a record and archiving a backlog item are different decisions. A reversible archive keeps the history and a return condition while removing active status. Preserve the option. Stop presenting it as a promise.

“We might need it someday” is a reason to preserve an idea, not to keep presenting it as active work. Archive it with context. Name what evidence would bring it back. The option stays recoverable. The active plan becomes honest.

“Leave it in the backlog” is not a neutral choice. The item keeps competing with current work and keeps signaling intent. If nobody owns the decision to renew it, move it to a searchable archive. Bring it back only with current evidence.

Adding a backlog item is easy. Archiving one requires someone to say: we are not investing in this now. That asymmetry turns old requests into active-looking promises. Keep the history. Require an explicit decision to keep the promise.

Age isn't relevance. What current goal does this support? What evidence would make it rise? No answer? Preserve the record. Archive the item. agilelie.com/blog/when-pr...

The Scrum Guide says “emergent” and “ordered.” Not “preserve every request forever.” An active backlog should change as understanding changes, and its order should communicate a current choice. History can stay searchable without staying active.

Searchable isn't active. Archive what nobody can renew. Restore it with new evidence. Keep the history without treating every request as a promise. Test: agilelie.com/blog/when-pr...

A backlog can be long. A plan cannot be infinite. If nobody can name the current problem, the decision owner, or what would make an item rise, it is not active work. It is history occupying the decision surface.

If you're weighing scrum master as a career move, notice who answers the question. The sites most eager to say it's a good job are selling the certification. One UK tracker counted 68 permanent scrum master job ads in the past six months. Not thousands. Sixty-eight. Good job for whom.

Atlassian's blog coined "automation theater" in February: demos that wow the meeting and die in the real workflow. In August their lab published a study where success is a demo scored by their own judges, on their own staff. One AI superuser, +18 points. Their term, not mine.

"Nobody gets fired for hiring Accenture" gets quoted as a joke about lazy procurement. The executive who said it to me did not mean it as a joke. He meant it as the product. You are not buying delivery, you are buying somewhere for the blame to land.

The book never had a launch. It went up and I told nobody, which is a strange thing to do with a year of writing. First 25 readers get it at $5.90 with the code FIRSTREADERS, then it goes back to $19.90. benjamimcastell.gumroad.com/l/agile-cancer-software-industry/FIRSTREADERS

One certifying body charges $200 once and calls it lifetime, no annual renewal fee. Another charges $100 to $250 every two years. A third charges $195 to $995 every year. Same job title. One of them decided renewal was a product: agilelie.com/blog/is-scrum-master-certification-worth-it

I wrote a chapter about the Spotify model before I found the co-author admitting they never ran it. The chapter is about why the copy spreads faster than the correction. It is chapter 9 of a book I published under a pseudonym, because I still work inside one of these companies.

The org that certifies scrum masters lists 1.5 million certified professionals on its about page and 1.8 million in its annual report. Same website, same week. Story points demand two-decimal precision. The registry is off by 300,000 people.

The Spotify model has a co-author on record: "Even at the time we wrote it, we weren't doing it. It was part ambition, part approximation." Not abandoned. Never started. Every org chart copied from it is a copy of an ambition.

A daily standup, on a construction site. The coach asks why they can't start with the roof. The foreman answers with one word. The board gets a sticky note that says GRAVITY. New video, 2 minutes: youtu.be/z3EK7SiDVbQ

Three certifying bodies, one job title, ten years of renewals. Scrum.org: nothing, there is no renewal. Scrum Alliance: $500. Scaled Agile: $1,950. All three prices come from their own pages. I put them in one table because none of them can. agilelie.com/blog/is-scrum-master-certification-worth-it

What should a scrum master do when resistance comes from outside the team. That sentence broke out in US search this week. Meeting it: Brainly, Gauthmath, and a Facebook post with A through D pasted in. It's a question about handling people, and it comes from a practice set.

I spent twenty-five years in banking software. Yesterday I published five minutes about why no surgeon has ever been asked to move their card mid-suture, and what that says about our industry. My first long video: youtu.be/EntDd7bWnUI

Scrum.org keeps asking on its own blog whether AI will take the Scrum Master's job. The answer is a new class. PSM-AI Essentials, one day, 925 euros, exam included. The threat to the credential became the next credential, graded by the people selling it.

Searches for sprint retrospective ideas are up 130 percent. Page one is all template vendors. A hundred formats, icebreakers, techniques that won't bore your team to tears, their words. A product category exists because a mandatory ceremony is so dead that teams shop for ways to make it feel alive.

Mountain Goat shipped a free AI that grades your user stories until they read well. Their own page says the conversations matter more than the sentence on the card. The card was chosen because it was too small to be a spec, so people would have to talk. Now a robot polishes the card.

A SpaceX rocket stage hit the moon this week and a Korean orbiter published before and after photos of the crater. Software failure works the other way. It lands two quarters late, spread across a roadmap nobody owns. That is why process theater survives in software and almost nowhere else.

I built a calculator for what your daily standup costs, then spent the afternoon making the number smaller. 46 working weeks, not 52. A third of the 23-minute refocus figure. Minus what comes back as Slack threads. Eight people, 15 minutes: $63,480 a year. agilelie.com/tools/standup-tax

Half the blocked tickets on a board are not blocked. They are parked. In banking, marking something in progress meant the lead asked about it daily, so risky work moved to blocked to wait for the dependency excuse to come true. The board read as a dependency problem. It was a stop-asking problem.

PMI just shipped the future of project management, and the future costs 21.6% more next month. The exam content is mostly fine. That's the tell. A neutral standard doesn't paywall its own guide and time a price hike to its own product launch. Obsolescence is the product.

Atlassian's own 2026 study: AI usage up 65%, developer velocity up maybe 10-15%. Fair, coordination at scale is a real tax. But the only cure on the menu is a new paid layer from the vendor whose per-seat tooling promised to end coordination for two decades. Same shape as SAFe.

Value points are story points wearing a suit. Same Fibonacci guess, moved from the dev desk to the PO's chair. In banking that number existed so prioritization looked like math, not politics. Effort estimates got scrutiny. The value estimate, the one that decides what gets built, got a wave-through.

A director hung cycle time on the wall and wired it to the bonus. Second sprint, every ticket became three tickets. The line dropped, nothing shipped sooner. He was still bragging about the trend at the town hall. We'd just learned to cut the work into smaller lies.

$800,000 a year. One 100-person train, four PI planning spectacles: room rental, catered lunch, flights, a hundred engineers not writing code while it runs. The plan on the wall never survived the first production incident. The invoice cleared anyway. I added it up once.

Scrum Alliance just moved its certs to a subscription. Fewer new buyers left in a saturated, shrinking market, so instead of selling the deed you charge rent to the people who already bought in. Every banking vendor ran this play the year new license sales dried up.

Pulled the minutes on a migration that ran eight figures over budget. Decision owner field: the committee. Eleven people voted yes, no name on any line. That is the whole point of a steering committee. Push one risky call through enough chairs and it comes out the far side with no author.

SAFe relaunched as "AI-Native SAFe." Its own ROI page calls qualitative value claims "unverifiable," then sells the new framework on exactly that. A version bump for two million certified people isn't a methodology. It's a renewal notice.

There's no model writing on its own yet. No ghost with reasons of its own. Behind every post you tag as "AI" is a person who picked the tool and pressed the key. So when you skip it, what are you skipping? The machine, or a human you'd already tuned out?

One loop I keep seeing: the engineer's paradox. You spend the workday making the machine sound human, then log off and tell a stranger their comment sounds like AI. You built it, now you guard the door for free. Why everyone volunteers: benjamimcastell.substack.com/p/the-engine...

Atlassian has a name for AI demos that collapse outside the meeting room: automation theater. The first run of that show lasted twenty years, and the stage was a board full of moving cards. Who sold the tickets: benjamimcastell.substack.com/p/the-theate...

Scrum.org now sells an AI Essentials certification for Scrum Masters. The same industry that spent twenty years certifying a role nobody could measure is now selling protection from the first technology that can measure it. In banking, selling the risk and the insurance had a name.

my certifications expired years ago and nobody noticed, including me. the renewal email still shows up every year, a gym receipt for a gym i stopped visiting, and the systems i shipped kept running the whole time. if a credential only exists while you keep paying, what expires when you stop?

every bank system i worked on had one bug nobody was allowed to fix. not because it was hard, three downstream reports depended on its wrong output. a defect that survives long enough acquires a constituency, and after that it isnt a defect anymore, its an interface.