Why Spotify's product management won't work for your business – even less so with AI agents

Why this article is back after three years

I published this article on pedalix.com in February 2023. Back then the question was: why do companies fail when they copy Spotify's squad model? Three years on I find the text has become more relevant, not less – for a reason nobody had on the radar in 2023. The most autonomous team member in your organisation is no longer a squad. It is an AI agent writing code on its own for hours.

The Spotify debate was never a debate about Spotify. It was a debate about how much autonomy an organisation can handle – and what must be in place before freedom produces speed instead of chaos. In 2026, exactly that question decides AI transformations. Hence this revised text: the same lessons, a new emergency.

The Spotify model in brief – and why it died at Spotify itself

As a reminder: squads are cross-functional, autonomous teams of up to eight people, responsible from idea to operation. Tribes group squads, chapters bundle expertise across teams, guilds are loose communities of interest. The idea behind it: community over hierarchy, local decisions over top-down approvals, trust over control. On slides it looked like the future of product organisation.

Reality was less comfortable. Henrik Kniberg and Anders Ivarsson, the authors of the famous 2012 whitepaper, described a snapshot – not a blueprint. Former employees such as Jeremiah Lee later publicly criticised the model for solving the wrong problems and presupposing a human maturity no org chart can produce.

The hardest design flaw: no engineering managers accountable for outcomes. In a disagreement, the product manager negotiated with every single engineer on the team – instead of with one technical leader. The result: lost time, endless back and forth, knowledge that never got shared. Spotify itself has long since moved on from the model.

And the main reason the copies failed has been in this text unchanged since 2023: you are not Spotify. A fast-growing, lavishly funded B2C player with hundreds of engineers and a single product can afford experiments that immediately create friction in a DACH mid-market company with a grown codebase, regulated customers and three parallel product lines. Context beats framework – always.

The three lessons that still apply in 2026

Lesson 1: autonomy needs alignment. Spotify's own retrospective coined the formula “minimum viable agility”: every new team reinvented the wheel, because maximum freedom also meant maximum arbitrariness. Autonomy only works inside a frame – a one-sentence mission, a shared priority logic, agreed interfaces. Freedom without a frame isn't speed, it's distributed waste.

Lesson 2: collaboration is a skill, not an org chart. Spotify had smart people – but smart is not the same as experienced. What was missing was a shared language for discussing process problems and the practice of measuring performance together. Fine-sounding titles replaced no shared practice. If you want cross-functional teams, you must actively build the shared language: shared terms, shared metrics, documented knowledge instead of insider knowledge in senior heads.

Lesson 3: ownership needs an address. Managers exist for a reason: someone has to own the outcome – not just career development. Cross-functional as the default, functional only where necessary, plus clearly defined, written and deliberately changeable roles: who leads technically? Who owns the roadmap? Whether you choose the triad model (engineering, product, design as equals) or a general manager per area is secondary. That it is answered is primary – especially when the organisation grows fast.

2026: your most autonomous team member is an AI

Now to why this text earned an update. Coding agents are the squad promise in extreme form: they work autonomously, make local decisions, need no approval from above – and they are faster and cheaper than any team Spotify ever assembled. Execution is no longer the bottleneck. The bottleneck isn't the coding, it's everything before it: overview, prioritisation, clean inputs.

And so the Spotify mistakes repeat at machine speed. An agent with a vague mission is a squad without alignment: it runs in the wrong direction for an hour and produces an hour of damage. An agent without a shared language guesses, because the domain knowledge sits scattered across five systems and senior heads – from conversations with product teams we know that is exactly where most planning time seeps away. And an agent without an ownership address is the old diffusion of responsibility, only worse: when nobody owns the outcome, nobody notices until production burns.

Automate bad input and you just ship the mess faster.

The answers are the same three lessons, translated into everyday agent work. Alignment now means: an agreed spec as the contract before the agent builds – the define gate. A shared language means: context searchable by human and machine alike – code, decisions and tickets as one connected index instead of twenty open tabs.

Ownership means: the human owns the outcome and checks at the gate, the agent executes – with narrow diffs and one intent per pull request. That is exactly how we built Teklens around the Discover, Define, Build, Operate cycle: human and AI in the same rhythm.

Recommendations

• Copy principles, not vocabulary. Squads, tribes and guilds are answers to Spotify's 2012 context. Adopt the principles behind them – small cross-functional teams, a clear mission, a limited blast radius – and name them the way your organisation speaks.
• Build alignment before autonomy. Mission, priority logic and ownership addresses first, freedom second – for teams and agents alike. The autonomy test above shows you where the gaps are.
• Onboard agents like new team members. Context, an agreed spec, review at the gate, a limited scope of effect. What you wouldn't let a new senior do – work unbriefed in the core system – you don't let an agent do either.
• Make knowledge infrastructure. The shared language doesn't emerge in a workshop but in searchable context: code, decisions and tickets connected – so human and agent find the same answers instead of repeating the same questions.

Scope & caveats

• The Spotify criticism rests on public accounts – the Kniberg/Ivarsson whitepaper (2012, explicitly meant as an example) and reports by former employees such as Jeremiah Lee (2020). Spotify as a company never responded comprehensively; treat the accounts as well-documented perspectives, not official record.
• “You are not Spotify” cuts both ways: the recommendations here are context-dependent too. Triad or GM model, degree of agent autonomy, depth of the gates – calibrate them to team size, regulation and the maturity of your codebase rather than to this article alone.
• The magnitudes on scattered knowledge and planning effort come from conversations with product teams in the DACH region, not from a controlled study – we deliberately label them as such until pilot metrics are available.

Matching use cases from the library

From the article straight into practice: these use cases put the concepts to work with Teklens.