Beyond the Sprint

If AI is rewriting how work gets done inside companies, why are we still managing it like it's 2005?

For the last nine months I've put all my attention on transforming companies with AI. Not running pilots. Not shipping apps. Transforming whole organisations. The problem is that most of them still manage work with methods built for a different bottleneck. Agile, Scrum, sprint planning, the whole stack assumed developer hours were the scarce thing. In the age of AI, everyone is a developer. That changes what the operating model has to do.

The Agile Manifesto was signed in 2001 by seventeen developers trying to fix broken software projects. Their four priorities weren't wrong. But what enterprises do in the name of Agile today, the manifesto's authors would not recognise.

"Responding to change" became an excuse to never finish anything. "Working software over documentation" became a quiet retreat from institutional knowledge. "Velocity" replaced quality. Story points, burn-down charts, and sprint reviews started measuring how fast a team was moving, not whether what they were producing was any good.

The dollar cost of this drift is substantial. McKinsey puts technical debt at 20 to 40 percent of engineering capacity in most large organisations. The Consortium for Information and Software Quality estimates that poor software quality cost US companies $2.41 trillion in 2022. The Standish Group found that in 2020, only 31 percent of software projects were considered successful. Agile has been the dominant methodology across most of the period that produced those numbers.

None of this means 2001's developers were wrong. The point is that the operating model an enterprise wraps around their principles has, over twenty-five years, calcified into a ceremony tax that the work no longer requires.

AI changed the bottleneck

The clearest summary of what's happening comes from a comment on a piece Chamath Palihapitiya published last year:

"Agile made sense when dev hours were the bottleneck. AI flipped it. Now the bottleneck is context. Better docs and dev infrastructure matter more than sprint velocity."

The data backs it up.

Anthropic's Claude Code team, by mid-2025, was shipping roughly 60 to 100 internal releases per day, with engineers averaging five pull requests per person per day. Boris Cherny, who founded the project, has reported that around 90 percent of the codebase is written by Claude Code itself. When the team doubled in size, throughput per engineer rose by 67 percent, the inverse of the dilution most scaling produces. The team uses no sprints, no story points, and no rigid backlog grooming.

Mike Krieger, Anthropic's Chief Product Officer, has been explicit about where the friction now lives. With AI writing the bulk of the code, the bottleneck has migrated upstream: to deciding what to build, aligning stakeholders, and pushing through merge queues. The role of engineering management is changing. The cost of "getting the requirements right" has not fallen at all, while the cost of writing code has fallen by an order of magnitude.

Other frontier organisations show the same pattern. Shopify's CEO, in a 2023 memo, deleted roughly 10,000 recurring meeting events and ten thousand hours of meeting time per week from the company's calendar. Two years later the same CEO mandated "reflexive AI usage" as a baseline expectation; headcount requests must now first prove that AI can't do the job.

Linear, which is widely admired for its product, runs sixty people across its entire company without sprints or story points. Their published method states that the goal is to "communicate using principles rather than guidebooks."

What these organisations have in common is small process surface and high autonomy at the unit of work, anchored by clear, measurable outcomes and trust-based relationships at the team and stakeholder level.

The ceremony tax is measurable, and growing

The overhead isn't theoretical. It appears in time-tracking exports, sprint reconciliation logs, and stakeholder interviews. A few consistent patterns:

Reporting overhead at the product-owner layer. A consistent figure from multiple enterprise engagements: fifteen to twenty hours per week per senior product person on admin and reporting work. Across a team with five product owners, that's 75 to 100 hours per week consumed by ceremony, not by the product. None of those hours produce a customer outcome.

Multi-system reconciliation at sprint boundaries. A typical enterprise sprint close requires reconciling at least five systems: the ticketing system, the time-tracking system, the client-facing project management surface, the invoicing instrument, and a delivery report. A sprint can't close until all five agree. The real-world cost on serious engagements ranges from twenty to forty hours per sprint of work that is not billable, not visible to the customer, and not reducible without changing the operating model itself.

Meeting overhead. Shopify's calendar purge removed roughly 76,500 hours of recurring meetings per week. Most enterprises that haven't done a comparable audit are running with the same overhead, hidden on the calendars of their senior people.

As AI lowers the cost of the work itself, the cost of the ceremony rises in relative terms. The same operating model, run on a faster substrate, produces more visible waste, not less.

What's emerging

The literature on what comes next isn't unanimous, but it converges. Six lines of work all point in the same direction:

Shape Up (Basecamp / Ryan Singer): six-week cycles instead of two-week sprints; appetite instead of estimates; a betting table where senior leadership commits to a small set of shaped projects each cycle. No standups, no story points.

Outcome-based product teams (Marty Cagan / SVPG): feature teams are given output (a roadmap of features to build); product teams are given an outcome (a customer or business problem to solve). Foundation models are crossing the threshold to act as scalable, always-on coaches, which removes one of the historical excuses for heavy management overhead.

North Star metric programs (Amplitude / Sean Ellis / Reforge): the discipline of identifying a single metric that best captures the customer value the product creates, and then orienting the entire organisation around moving it. Cadence is whatever the input demands, not a fixed two-week heartbeat.

Amazon's Working Backwards: each significant initiative has one owner accountable end-to-end, a press release written before the work starts, and a FAQ that anticipates the customer's actual experience. Sprint reviews are replaced by six-page narrative documents.

Consultancy convergence (2025): McKinsey, BCG, Deloitte, and Bain all published 2025 AI-transformation reports. None of them recommends Scrum. All four converge on senior-leadership-owned governance, cross-functional oversight, and outcome measurement as the differentiator between leaders and laggards.

Six lines of independent work, each from a different vantage, all describe operating models that are smaller in process surface, more outcome-anchored, more trust-based, and more senior-talent-leveraged than enterprise Agile.

Where this leads

From the convergence above, four principles are sufficient to redesign a client engagement:

1. Outcome over output. Each consultant operates against one named outcome that the client cares about and the consultant can move. The dashboard is the canonical surface. It replaces the sprint review, the status meeting, and the weekly report.

2. Trust over surveillance. The retainer replaces hourly billing. Per-ticket time tracking, sprint-level reconciliation, and the multi-system audit trail go away. What survives is a written monthly outcome review and a weekly written summary.

3. Collaboration over coordination. Each consultant has one named counterpart inside the client organisation. That person is empowered to make the decisions the work requires. Coordination between consultants happens consultant-to-consultant, not via a central project management office.

4. Honesty about what survives. This is not "no process." Three things survive intact: contractual rigour in the form of clear outcomes and a written master agreement; change-order discipline when the outcome scope changes materially; and an audit trail in the form of the monthly review, the weekly check-in, and the dashboard.

The full paper goes through what this looks like in practice: the pricing structure, the cadence, the exact shape of what dies and what survives, and what each side gives up in the trust exchange. If you work in consulting, you're a product leader who's tired of Jira hygiene consuming your senior staff, or you're thinking about how to restructure your engagements for the AI era, it's worth the read.