The rise and fall of Agile: why Scrum changed the industry — and what’s left of it in 2026

I was in the room when Scrum felt like a jailbreak. We had been shipping under a Gantt chart that lied every month. Two-week sprints, a backlog, a standup that was actually about blockers — that was better than a status meeting with twenty people and a projector. Then we did it for a decade and the jailbreak became the jail: story points as a performance metric, a sprint commitment that was a hostage note, and a “ceremony” calendar that left three hours a day to write code.

Agile did not fail as a set of values. Scrum changed the industry because it was a product you could buy: certified trainers, Jira workflows, a vocabulary executives could repeat. What is left in 2026 is a split. Some teams kept the feedback loop and threw away the costume. Some teams kept the costume and lost the loop. I have been on both. Here is the history I lived, and what I would still run.

Why Scrum won the 2000s and 2010s

Waterfall-as-practiced was a requirements document, a long build, and a surprise at UAT. Scrum offered a ritual that forced a demo. That demo was the revolution. Stakeholders saw a thing. The thing was wrong. You found out in two weeks instead of nine months. I will defend that until I retire.

It also offered roles that mapped onto existing power: Product Owner as the person who could say no (or the person who could not, depending on the company), Scrum Master as a facilitator or as a project manager with a new certificate. The 2001 Agile Manifesto was fuzzier. Scrum was a playbook. Consultants can sell a playbook. They cannot sell “individuals and interactions” without a slide that turns it back into a playbook.

Jira, VersionOne, Rally — the tools froze the playbook. A story has points. A sprint has a burndown. A board has columns that look like Kanban but reset every other Friday. Once the tool is the process, the process stops being a hypothesis. It becomes compliance. I have watched a team spend a morning making the burndown look healthy. That morning was the fall, not the rise.

What actually died

The idea that a two-week box is a unit of planning for every kind of work. Platform work, research, and a migration do not pretend to be a user story with acceptance criteria that fit a sprint without lying. We lied. We split fake stories. We called a spike a story. We carried tickets. Leadership called it “commitment.” I called it theater.

Story points as a forecast that became a quota. Fibonacci was supposed to be relative. Then someone computed velocity as a KPI and compared teams. I have sat in those reviews. They produce sandbagging and unsafe shipping. I still estimate when a conversation about size is useful. I do not publish a velocity target.

The Scrum Master as a meeting chaperone. The useful version is someone who removes a blocker and protects a team from a drive-by scope dump. The common version is someone who runs the clock and updates the board. In 2026 a lot of companies quietly renamed the role or folded it into an EM. That is not a tragedy when the EM actually unblocks. It is a tragedy when nobody unblocks and we just have fewer certificates on the wall.

SAFe and the other scaling frameworks as a default. I have been inside a PI planning that filled a hotel. We aligned. We also spent a week not shipping. For a 2,000-person IT org with regulatory trains, a scaled framework can be the least-bad bus schedule. For a 30-person product company it is a way to import a bureaucracy you did not earn. I will not pretend SAFe is Agile. I will pretend it is a governance product and evaluate it as one.

What I still steal

A regular demo to people who can change their minds. Cadence matters. Two weeks is fine. One week is fine for a small team. A month is fine if the work is fat. The demo is not optional. If you cannot demo, you do not have a increment. You have a hope.

A backlog that is ordered, not a wish list with 400 tickets. The PO or the PM or the founder — someone with a face — says what is next. The team says how. When those two roles collapse into “leadership wants everything,” you do not have Scrum. You have a queue with standups.

A retrospective that can change a rule. If the retro cannot kill a meeting or change a deploy policy, it is a feelings circle. I will keep retros that produce one experiment. I will cancel retros that produce a Miro of adjectives.

WIP limits. That is Kanban, and I like it more than sprint commitments for interrupt-driven teams (SRE, a platform group, a support-heavy product). A board that shows we already have too much started is more honest than a sprint that “committed” to twelve items and finished five.

What I would run on a real team in 2026

I would start with outcomes and a weekly demo, not with a framework name. I would keep a single ordered list. I would plan in a conversation when the work is uncertain — a day of design, not a two-day PI. I would use sprints if the team likes the heartbeat and if we treat the end of the sprint as a review, not as a contract with finance.

I would measure lead time and how often we can release, not points. I would keep on-call and product work on the same board so we stop pretending interrupts are free. I would staff so that a vacation does not freeze a “squad” that was already a single point of failure with extra standups.

I would not ban Agile words. I would ban the ones we cannot define. If someone says “we need to be more Agile,” I ask: which feedback loop is too slow, and what are we willing to stop doing to speed it up? If they cannot answer, we are buying a training, not a change.

Remote 2026: a daily 10-minute standup is optional if the team writes blockers in Slack and the EM reads them. A 45-minute standup that is a status tour is a tax. I have killed those. The work got faster. Nobody missed the tour except the person who used it to be seen.

Coaches, managers, and the meeting I will still interrupt

I have been the manager in the room when a coach tried to save a sprint by redefining “done.” Done became “dev complete.” QA became next sprint. Production became a hope. That is how you get a board that is green and a customer who is not. I will interrupt that meeting. Done means a user can touch it, or we write “not done” and we stop scoring the points. The coach who helps you say that is worth a contract. The coach who helps you hide that is a cost.

I have also been the engineer who used Agile language to avoid a date. “We’ll iterate” is not a plan for a compliance deadline. Some work has a hard external clock. Scrum does not repeal the clock. It can still slice a path to a thinner legal yes. If we cannot name the thin yes, we are not being Agile. We are being late with better sticky notes.

On a team I run now I keep a written working agreement that is one page: how we pick work, how we review, how we ship, how we page. We change it in the retro. We do not need a 40-page playbook. We need a page we actually open. That is what is left of the methodology I can still respect — a local rule set with a kill switch, not a brand.

The industry leftover

Job listings still say Scrum. Consultants still sell Scrum. Jira still looks like Scrum. That leftover is real. I will speak the language in an interview. I will not install the full costume on a team of eight that already ships weekly. The rise was a better demo. The fall was a better certification. What is left worth keeping is the demo, the ordered list, the right to change the process when the process is the bottleneck. The rest is a museum I can visit without living there.

I have used Shape Up for a season when the work was project-shaped and the appetite was a six-week bet. I liked the appetite conversation. I did not like pretending every problem is a six-week bet. Mix the tools. Do not join a new church.

I still like the Manifesto’s bias toward working software and toward people who can talk. I do not like the industry that turned that bias into a two-week theater with a burndown. In 2026 I will take the bias and leave the theater. If a team wants to call that Scrum, they can. I will be looking at whether a user saw a change, not at whether the points closed.