Last week, we covered the India AI Impact Summit — $200 billion in infrastructure commitments, the Delhi Declaration, every major AI CEO on the same stage. The macro picture is clear: the investment is flowing, the governance frameworks are converging, the infrastructure is being built.
This week, we shift to what matters more: what's already changing inside companies right now.
Because while the world was watching heads of state in New Delhi, something quieter and arguably more significant was happening inside one of the world's largest technology companies. And it has direct implications for every enterprise deploying AI.
Spotify's Engineers Have Stopped Writing Code
During its fourth-quarter earnings call on February 12, Spotify co-CEO Gustav Söderström told investors that the company's most senior engineers — the best developers they have — haven't written a single line of code since December. They generate code using AI and supervise the output.
This isn't a pilot. It isn't a proof of concept. It's how Spotify builds software now.
The system behind this shift is called “Honk” — an internal orchestration platform built on Anthropic's Claude Code. It allows engineers to issue natural-language instructions from Slack on their phones and receive deployable code in return. An engineer on their morning commute can tell Claude to fix a bug or add a feature, receive a test build, provide feedback, and have the updated version merged into production — all before arriving at the office.
The results are measurable. Spotify shipped more than 50 new features and changes throughout 2025, including AI-powered Prompted Playlists, audiobook Page Match, and About This Song. The company credits this velocity to AI-assisted development. Premium subscribers grew 10% year-over-year to 290 million. Monthly active users reached 751 million. Revenue grew 13% to €4.5 billion. Gross margin improved to 33.1%.
Söderström was direct about the implications: “There is going to have to be a lot of change in these tech companies if you want to stay competitive, and we are absolutely hell-bent on leading that change. It will be painful for many companies, because engineering practices, product practices, and design practices will change.”
He also acknowledged the instability: “The tricky thing is that we're in the middle of the change, so you also have to be very agile. The things you build now may be useless in a month.”
Why This Matters More Than Summit Announcements
Summit commitments are measured in years and billions. What Spotify demonstrated is measured in weeks and features shipped. The difference matters because it reveals the operational reality that enterprises face right now — not in 2028 when Adani's data centres come online, but today.
Here's what Spotify's transformation actually tells us about where enterprise AI has arrived.
The Developer Role Is Being Restructured — Not Eliminated
Spotify's senior engineers didn't lose their jobs. They changed what they do. The shift is from writing code to architecting solutions, supervising AI output, reviewing for quality and security, and making high-level design decisions. The engineering role has moved from implementation to orchestration.
This mirrors what Deloitte found in its 2025 study: only 11% of organisations use agentic AI in production. Spotify is in that 11%. But the shift it's demonstrating — engineers as supervisors of AI-generated code rather than authors of manual code — is the trajectory for every technology organisation.
The critical nuance: this works at Spotify because they invested years in infrastructure that makes AI-assisted development possible. Since 2020, Spotify has used “Backstage,” an internal developer portal where every team can track code ownership and dependencies. Since 2022, they've used “Fleet Management” — a framework that performs code changes across hundreds of repositories simultaneously. Honk and Claude Code sit on top of this infrastructure, not in place of it.
Enterprises that try to replicate Spotify's results by simply giving developers AI coding tools without the underlying infrastructure — the dependency mapping, the automated testing, the deployment pipelines, the governance frameworks — will get AI-generated code they can't trust, can't deploy, and can't maintain.
The Talent Pipeline Is Already Shifting
The transformation isn't limited to how engineers work. It's changing who becomes an engineer.
Computer science enrollment declined 6% across the University of California system in 2025, following a 3% decline in 2024 — even as overall college enrollment rose. Students are shifting to AI-specific programmes. UC San Diego, MIT, University of South Florida, and University at Buffalo have all launched or expanded dedicated AI degrees that are attracting thousands of students.
This isn't students abandoning technology. It's students recognising that the skills that matter are changing. Traditional software engineering — writing code, debugging implementations, managing deployments — is being augmented and in some cases replaced by AI systems. The emerging skill set centres on AI orchestration, prompt engineering, system architecture, and quality oversight.
For enterprise leaders, this has hiring implications that are already urgent. The engineers you recruit in the next 12 to 18 months will work in fundamentally different ways than the engineers you recruited two years ago. Job descriptions, interview processes, onboarding, and career paths all need to reflect the shift from code-writing to AI-supervising.
AI Is Restructuring How Customers Find You — Not Just How You Build
While Spotify was transforming how it builds software, LinkedIn was transforming how it measures success — because AI search fundamentally changed how its content gets discovered.
LinkedIn reported that non-brand, awareness-driven B2B traffic declined by up to 60% as AI-powered search experiences reduced clickthrough behaviour. Rankings stayed stable — but clicks collapsed. People were getting answers from AI-generated summaries without ever visiting the source.
LinkedIn's response was to abandon traditional SEO metrics entirely and shift to visibility-based measurements centred on mentions, citations, and presence within AI-generated responses. A cross-functional task force developed new optimisation guidance tailored to what they call “generative engine environments.”
Research analysing 76,000 websites confirms the pattern: ChatGPT processes billions of daily prompts but drives dramatically less referral traffic than Google, because it resolves queries within the interface rather than sending users to external sites.
For enterprises, this is a double restructuring. AI is changing how you build products (Spotify) and how customers discover them (LinkedIn). Both transformations are happening simultaneously. Both require different strategies, different metrics, and different organisational capabilities than what most enterprises currently have.
The Pattern Beneath the Headlines
Zoom out from any single story and the pattern is clear: AI isn't arriving at enterprises in the future. It's restructuring how they operate right now, across multiple dimensions simultaneously.
How software is built: Engineers supervising AI-generated code rather than writing it manually. Internal orchestration platforms (Spotify's Honk) built on top of AI coding tools (Claude Code). Feature velocity accelerating dramatically.
How talent is developed: CS enrollment declining as AI-specific programmes grow. The skills that matter are shifting from implementation to orchestration, from coding to architecture and supervision.
How products are discovered: AI search absorbing traffic that used to flow through traditional channels. Visibility within AI-generated responses replacing clickthrough as the primary metric.
How competition is structured: Meta's $2 billion acquisition of Manus in December — buying an AI agent startup with $100 million in annual revenue and 80 million virtual computers deployed — signals the shift from foundation model development to deployable AI agents that execute real work. The race has moved from “who builds the best model” to “who deploys the best agents.”
How geopolitics shapes technology choices: Microsoft's Brad Smith warned at the India AI Summit that American tech companies should “worry a little bit” about Chinese government subsidies for AI competitors. Gartner's Julian Sun described the global AI landscape becoming “multi-polar” across different layers of the tech stack. The model-agnostic architecture we've advocated since blog one isn't just a technical preference — it's a geopolitical necessity.
What Enterprise Leaders Should Do This Week
The summit gave you the macro picture. Here's what the micro picture demands.
Audit your development workflows. Ask your engineering leads: what percentage of new code is AI-generated versus manually written? If they don't know, that's the first problem. If it's less than 30%, you're falling behind the trajectory that companies like Spotify are on. If it's more than 50%, ask what governance, testing, and review infrastructure supports it.
Assess your AI infrastructure readiness. Spotify's success with AI-assisted development rests on Backstage (dependency mapping), Fleet Management (cross-repository automation), and years of platform engineering. AI coding tools without this infrastructure produce unreliable results. Evaluate whether your development environment is ready to support AI-assisted workflows — or whether you need to build the foundation first.
Rethink your hiring strategy. The developer you need in 2026 is someone who can architect systems, supervise AI output, and make high-level design decisions — not necessarily someone who writes the most elegant code by hand. Update your job descriptions, interview processes, and evaluation criteria accordingly.
Measure AI's impact on your customer discovery. If you haven't assessed how AI search is affecting your traffic, leads, and customer acquisition, do it now. LinkedIn's 60% decline in awareness-driven B2B traffic is not an anomaly — it's the beginning of a structural shift. Your marketing and content strategies need to account for a world where AI-generated answers compete with (and often replace) direct website visits.
Build for multi-polar. The geopolitical dimension of AI — Chinese subsidies, European regulation, Indian infrastructure, Gulf sovereign frameworks — means that enterprises operating across markets need technology architectures that work across multiple AI ecosystems. Model-agnostic isn't a feature. It's a requirement.
The Transition
For three weeks, this blog series tracked the forces reshaping enterprise AI from the outside in — model competition, platform consolidation, infrastructure investment, governance frameworks, and the India AI Summit's $200 billion declaration.
Starting this week, we turn the lens inward. The external forces are clear. The question now is execution: how enterprises actually implement, deploy, and govern AI systems that deliver measurable business outcomes.
Spotify didn't wait for the summit to build Honk. LinkedIn didn't wait for the Delhi Declaration to restructure its discovery metrics. Meta didn't wait for a governance framework to buy Manus for $2 billion.
The enterprises that will lead aren't the ones that understood the summit best. They're the ones that are already building — and have been since before anyone in New Delhi picked up a microphone.