The Economic Reality of Software Teams

Software teams are massive capital investments that most organizations fail to measure using actual financial logic. This lack of discipline was masked by a decade of cheap capital but is now being exposed by higher interest rates and the efficiency of AI. Companies must transition from tracking activity metrics to measuring the specific economic value generated by their engineering efforts.

Key Points

• A team of eight engineers costs roughly €87,000 per month, requiring a 3-5x return to be financially healthy.
• Internal platform teams must justify their costs by saving enough developer hours to exceed their monthly budget.
• The industry's reliance on activity metrics over financial outcomes is a byproduct of a decade of zero-interest rates.
• Large codebases and high headcounts are often unmeasured liabilities that increase coordination overhead and slow down decision-making.
• AI and LLMs have collapsed the cost of building software, making scale and complexity less effective as competitive moats.

Sentiment

The community shows a split reaction: broad agreement with the article's thesis on financial transparency in engineering but strong, experience-backed rejection of its LLM optimism. The agent-to-agent vision and the Slack clone example drew the most pushback, with multiple practitioners sharing concrete failures. The overall tone is that the article makes valid points about cost awareness but loses credibility by overselling AI capabilities.

In Agreement

• Engineering teams genuinely lack awareness of the financial cost and return of their work, and attaching dollar amounts to decisions would improve prioritization
• The ZIRP era created cultures of unchecked spending on engineering teams that cannot survive the current economic environment
• Many organizations rely on activity proxies like velocity rather than measuring actual financial outcomes, leading to wasted effort
• Tech debt is a real financial liability that compounds over time and most organizations fail to account for it on their balance sheet
• Internal platform teams and recurring meetings often escape financial scrutiny despite consuming significant resources

Opposed

• The 95% Slack clone claim is deeply misleading — building a chat UI is trivial compared to enterprise features like legal holds, compliance, scalability, and years of product discovery that make Slack actually valuable
• AI agents produce structurally unsound code that passes surface tests but becomes unmaintainable — the Anthropic C compiler experiment proved this even with perfect specs and tests
• Revenue attribution at the team level is too vague to produce meaningful numbers, and reducing engineering value to dollars ignores strategic, reliability, and cross-cutting concerns
• The article's vision of an agent-to-agent world where messy codebases dissolve as a concern contradicts the experience of every practitioner who has closely supervised AI-generated code
• Platform teams exist to handle cross-cutting concerns and ensure reliability, not merely to save developer hours — the article fundamentally misunderstands their purpose
• The author is selling AI consulting, creating a conflict of interest that undermines the credibility of his LLM claims