Crashes, No Sound, or Screen Glitches?
Random freezes, missing sound and display glitches usually trace back to one bad driver. Find and replace yours safely.Free scan · under a minutePC Slower Than It Used to Be?
A free scan shows the junk files, broken settings and background clutter dragging Windows down - then fixes them in one click.Free scan · Windows 10 & 11Some links on this page are affiliate links: if you buy through them we may earn a commission, at no extra cost to you.
A February 2026 report says Spotify co-CEO Gustav Söderström told analysts that the company’s strongest engineers “have not written a single line of code since December.” The report describes an AI-assisted workflow built around an internal system called Honk. That is a striking claim about how some engineers may be implementing software—not proof that AI has replaced software engineers or that Spotify’s developers have stopped doing technical work.
The original earnings-call transcript or recording was not available in the reporting reviewed, so the quote and details about Honk should be treated as reported claims, not independently verified specifications. The report does not clarify whether “December” means a particular date or a continuous period; in this context, it appears to refer to December 2025.
What Spotify reportedly said
According to a report published February 14, 2026, Söderström made the statement during Spotify’s fourth-quarter earnings call. The report says the company’s strongest engineers had not written a line of code “since December.” It also describes Spotify using an internal system named Honk, with Anthropic’s Claude Code involved in generating, testing, and deploying software. The report says engineers can interact with the workflow remotely, including through Slack. Those details come from secondary coverage; the report does not establish how many engineers use Honk or how its production permissions work. Read the report.
The same report says Spotify released more than 50 features and improvements during 2025 and links its development pace to AI-assisted work. It provides no controlled comparison, engineering-hours baseline, defect data, or other method for measuring how much AI contributed. The figure is not, by itself, evidence that AI caused the output or made the company more productive.
#1 Best Overall
What “not writing code” does—and does not—mean
Software development includes more than typing implementation code. An AI agent may produce or modify files while people still define the task, decide what the system should do, assess the proposed change, and take responsibility for its consequences. The reported statement most directly concerns manual code entry.
| Claim | What the available account supports |
|---|---|
| Some Spotify engineers stopped manually typing code for a period | Reported by secondary coverage; the original call was not independently verified here. |
| AI is involved in implementation | The report describes AI-assisted code generation, testing, and deployment through Honk. |
| AI independently designs, secures, and operates Spotify’s software | Not established. The report does not explain the human approval gates or agent permissions. |
| Spotify’s developers—or most of them—no longer do engineering work | Not established. The statement refers to the company’s “strongest engineers,” without a headcount or definition. |
| AI caused a measured productivity gain | Not established by the figures or methodology in the report. |
| The same approach works for typical development teams | Unknown; Spotify’s reported experience is not a representative sample. |
How an AI-assisted engineering workflow can work
In an agent-based workflow, a developer gives a tool a task and relevant constraints. The agent may inspect a repository, edit several files, run commands or tests, and present a proposed change. A human still needs to judge whether the task was understood, whether the change fits the system, and whether its behavior is safe in production.
Implementation moves from typing toward direction and review
Engineers can break work into bounded tasks, specify expected behavior, supply examples, and set acceptance criteria. They may then review the generated diff, investigate failures, add or revise tests, and decide whether the change is ready to merge. If agents can work concurrently, coordinating their tasks and resolving conflicting changes also becomes part of the job.
Tests and deployment are not proof of correctness
Automated tests can catch many regressions, but a passing test suite only shows that the code passed the checks that were run. An agent can produce a change that satisfies existing tests yet violates an unstated business requirement, misses an edge case, or creates a security problem. Production readiness also involves deployment controls, monitoring, rollback plans, and responsibility for incidents.
What is known about Honk
The available account describes Honk as an internal Spotify system for natural-language requests and AI-assisted code generation, testing, and deployment, with remote interaction including Slack. It does not provide a technical design, explain which parts are handled by Claude Code or other components, or state whether agents deploy directly or require human approval. It also does not document how Spotify isolates code execution, protects credentials, logs actions, or rolls back changes. Those details matter to understanding the system, but they cannot be inferred from the reported headline.
Why experienced engineers may be first to stop typing
Senior engineers are often better positioned to delegate implementation to an agent because they know the system’s architecture, conventions, and failure modes. They can define narrower tasks, provide useful context, and recognize when a plausible-looking change is wrong. They are also more likely to know which tests, operational signals, and reviewers can validate a change.
Rank #3
That creates a distinction between automating code production and automating engineering judgment. If AI generates more of the implementation, the value of people who can specify, review, and own the result may rise. The reported claim does not show that those responsibilities have disappeared at Spotify.
Quick wins for a faster PC:
Repair Windows errors before they cause bigger problemsFix Now →Scan for outdated or missing drivers - takes under a minuteDriver Scan →Clear out junk files and repair common Windows errorsFree Scan →Anthropic describes Claude Code as an agent that can work in a codebase, read and modify files, run commands, and use integrations such as GitHub and MCP tools. It is offered through terminal, IDE, web, mobile, and Slack interfaces. These are product capabilities, not independent confirmation of how Spotify configured Honk. Anthropic’s Claude Code overview.
Does the claim show that AI makes developers more productive?
Not on its own. “No lines written by hand” measures neither the amount of useful work completed nor the cost and quality of that work. To judge productivity, a team would need to compare outcomes such as time from task assignment to production, human editing and review time, defects, security findings, rollbacks, maintenance effort, and customer impact.
A faster first draft may not shorten the full delivery cycle if review becomes harder or the change requires extensive correction. Conversely, an agent might save time on routine implementation while a human spends more time on architecture or validation. The relevant measure is the whole workflow, not the quantity of generated code.
Why Spotify’s experience may not generalize
The statement concerns a selected group described as Spotify’s strongest engineers, not a representative set of developers. The report does not define that group or say how many people it includes. It also does not establish that a team with different tools, code quality, or constraints would get the same result.
Do these 3 things before closing this tab:
1Clear out junk files and repair common Windows errors2Scan for outdated or missing drivers - takes under a minute3Repair Windows errors before they cause bigger problems- Testing and tooling: Mature automated tests, reliable continuous integration, clear code ownership, and standardized deployment make it easier to catch bad changes and run agent workflows.
- Repository and domain knowledge: An agent is more useful when the codebase and task are understandable. Undocumented behavior and changing requirements leave more room for plausible but incorrect output.
- Security and regulation: Some teams cannot send source code or data to an external model, or must tightly restrict what tools can access.
- Review capacity: Generating more changes can shift the bottleneck to reviewing and maintaining them. A small team may not have the specialist reviewers or platform support available to a large engineering organization.
- Operational safeguards: Remote or chat-based requests need strong authentication, least-privilege access, audit trails, isolated execution, and clear approval and rollback rules.
Risks a team should manage before delegating production work
- Incorrect scope: An agent can edit the wrong files or misunderstand repository conventions. Keep tasks bounded and inspect the diff before approval.
- Weak tests: Generated tests can repeat the implementation’s mistaken assumptions. Check that tests exercise requirements and meaningful failure cases.
- Security exposure: Repositories, issues, documents, and dependencies can contain malicious or misleading instructions. Treat that content as untrusted, limit credentials, and prevent agents from gaining broader access than a task requires.
- Hidden operational changes: Dependency, configuration, or permission changes may carry risks beyond the visible feature. Review them explicitly.
- Unclear accountability: Assign a human owner for each change, deployment, and incident; an agent cannot take organizational responsibility.
- Accumulating maintenance costs: Measure whether generated code is understandable and maintainable, not just whether it passes the immediate checks.
- Changing model behavior and usage costs: Model updates can affect results, while limits or usage-based charges can make long-running work less predictable. Track agent actions, model versions, and costs where possible.
What this could mean for junior developers
If AI agents take on routine implementation, one possible consequence is fewer entry-level tasks through which new developers learn a codebase by making changes, debugging them, and seeing how they behave. That is a concern, not an outcome established by Söderström’s reported statement: the account supplies no Spotify hiring, staffing, or career-progression data.
Teams that automate routine coding may need to make learning more deliberate, with supervised implementation, code-reading, debugging, testing, and increasingly complex ownership. The risk is that junior engineers are expected to review output before they have had enough practice to understand it; the opportunity is that they can learn to use agents while still building the fundamentals needed to check their work.
How another team can evaluate an AI coding workflow
A team does not need to begin by granting an agent production access. A controlled pilot can reveal where it saves time and where it adds review or maintenance work.
- Choose a bounded task: Start with a change that has clear acceptance criteria and a way to verify the result.
- Run in an isolated environment: Use a branch or sandbox with only the permissions and data the task requires.
- Keep human review in the merge path: Require a named engineer to inspect the diff, tests, dependencies, and configuration changes.
- Protect production systems: Use branch protection, least-privilege credentials, deployment approval, monitoring, and a rollback path.
- Record the full effort: Track agent time and usage, human prompting and editing, review time, test results, and downstream fixes.
- Compare quality as well as speed: Look at defects, security findings, rollbacks, maintenance burden, and customer outcomes alongside cycle time.
- Expand only when evidence supports it: Increase task complexity or autonomy gradually, and revisit permissions and safeguards as the workflow changes.
AI coding tools to evaluate
These products can help teams test different agent workflows, but buying a tool does not reproduce Spotify’s reported Honk setup. Internal orchestration, repository access, test quality, deployment controls, and engineering practices may matter as much as the model.
| Tool | Workflow fit | Price and qualification |
|---|---|---|
| Claude Code | Repository-level agent for terminal, IDE, web, mobile, and Slack workflows; Anthropic says it can read and edit code, run commands, and use integrations. | Anthropic’s pricing page observed August 18, 2026 listed Pro at $20 monthly or $17 per month equivalent with annual billing ($200 paid upfront); Max from $100 monthly; Team standard at $20 per seat monthly with annual billing or $25 monthly; Team premium at $100 per seat monthly with annual billing or $125 monthly; and Enterprise at $20 per seat monthly plus usage at API rates. Taxes and usage limits apply; prices may change. Pricing details. |
| Cursor | AI-enabled editor with agent features, frontier-model access, MCPs, skills, hooks, cloud agents, and usage-based Bugbot features listed on its pricing page. | Cursor’s pricing page observed August 18, 2026 listed Individual Pro at $20 per month and Teams at $40 per user per month. Higher individual tiers are also listed; check the current page for terms and limits. |
| OpenAI Codex | An alternative coding-agent product to evaluate, particularly for teams already using OpenAI’s developer ecosystem. | A standalone current Codex price was not stated in the product information reviewed. Check the applicable current pricing before comparing costs. |
The distinction behind the headline
Spotify’s reported claim is best read as a possible shift away from manually typing implementation code among some highly experienced engineers. It does not establish that AI has taken over the full engineering function, that the workflow has produced a measured productivity gain, or that other organizations can reproduce it. The key question is not only how much code an agent can generate, but whether people and systems can specify, verify, secure, and maintain what it produces.
Quick Recap
Product prices and availability are accurate as of the date/time indicated and are subject to change. Any price and availability information displayed on Amazon at the time of purchase will apply.
