Fumble of the decade

Heroku’s “magic” developer experience—spinning up production-ready apps by linking code to GitHub—has ended, with Salesforce shifting the platform into a maintenance-and-sustaining mode that prioritizes stability and security while cutting staff. The practical takeaway is blunt: existing paying customers can keep using Heroku, but new enterprise contracts won’t be offered, and the platform’s long-term trajectory is toward eventual disappearance, making migration a near-term risk-management task.

The collapse traces back to a business decision that undercut Heroku’s defining feature: a generous free tier. After Salesforce acquired Heroku, the free tier became a “money pit” rather than a customer funnel, and it was turned off—an action compared to a major consumer platform like Twitter removing its free access. That change, paired with later layoffs and stalled product momentum, left Heroku unable to compete with the modern onboarding expectations of developers who now expect fast experimentation, smoother deployments, and feature parity with newer platforms.

The transcript frames leadership churn as a key accelerant. In 2024, Heroku’s CEO Bob Weise opened a role for a chief product officer; the position appears to have remained unresolved long enough to coincide with further departures. Weise left Salesforce shortly afterward, and within a tight window both CEO and CPO losses are portrayed as a recipe for organizational drift—teams reportedly continued building without clear product direction, and modernization work that might have restored the free tier and improved developer-facing capabilities was effectively abandoned.

Against that backdrop, the discussion pivots to a broader question: can developers trust free tiers, and what should replace Heroku? The answer depends on how the platform charges. Server-based hosting reserves capacity 24/7, so “free” access is structurally expensive and vulnerable to shutdowns. Serverless platforms, by contrast, typically cost money only when requests arrive. That’s why platforms like Vercel (and similarly Cloudflare and Netlify) can offer generous free tiers: no traffic means near-zero infrastructure cost, and preview environments can be cheap because they’re mostly stored code that activates only when accessed.

The transcript also distinguishes between serverless and “real server” platforms. Railway provisions servers that often run continuously (even if they can sleep), which raises baseline costs and makes free tiers harder to sustain—yet Railway’s shift toward running its own infrastructure is presented as a reason its free tier became more stable and transparent. PlanetScale is treated as the cautionary tale: its free tier was cut quickly after it effectively mirrored paid plans by running multiple replicas, creating high costs without delivering the kind of customers it targets.

The migration guidance is direct. If a project can go serverless, the recommendation is to prefer Vercel; Cloudflare is acceptable despite weaker developer experience; Netlify is praised but not considered meaningfully different. For VPS-style hosting, Railway and Render are the safer picks, while other VPS options carry “crazy risk.” The closing message: Heroku is still technically alive for credit-card customers, but its direction is toward the drain—so developers should start moving now rather than waiting for the next policy change from Salesforce.