↓ Ir para o conteúdo principal

← todas as notas

📎 Webclip

How to waste half a million dollars

The article describes how startups drift from their true mission of building and selling a great product into running infrastructure. Using a fictional company, GoCo, it shows how moving from a platform like Heroku to self-managed servers appears to cut the hosting bill but actually consumes valuable engineering time, leads to new operational headcount, and can add up to roughly half a million dollars per year. The author argues that teams rarely account for these hidden costs or question whether operating servers aligns with the company’s mission.

Reading notes
#

  • The author says every startup they worked for effectively had the same mission: build a product customers love enough to recommend, charge for it, and make money.
  • The piece focuses on how teams get distracted from this mission and end up working on activities that do not help it.
  • The fictional startup GoCo starts by hosting its product on Heroku, which handles updates, patching, and general server concerns.
  • Early on, GoCo uses the platform to build a product, deliver demos, gain first users, and generate interest at meetups and conferences.
  • As costs grow, GoCo’s CFO reacts to spending “hundreds of dollars” on Heroku and pushes for running their own servers to lower the bill.
  • A top engineer, Grace, spends several weeks migrating the application to low-cost virtual machines, and the visible hosting bill goes down.
  • The author points out that this is a mistake because the company paid Grace for weeks of work, and those labor costs are usually ignored when celebrating a lower service bill.
  • During the migration, Grace could have been building new product features or optimizations that support the company’s mission.
  • The article emphasizes that missing features do not show up on a balance sheet, but without compelling features and a better product the startup cannot make money.
  • The author uses a stark example: the most effective way to cut costs is to fire everyone and shut down, which no mission statement endorses.
  • Over time, GoCo’s bargain servers experience downtime, especially during a critical conference, because patching and configuring is now the company’s responsibility.
  • Grace is not a full-time operations engineer and is likely evaluated on feature delivery, making server management a misalignment with her role.
  • After an outage, executives hold a post-mortem and decide they need an operations specialist or team to handle servers.
  • The discussion may cover the new ops salary but usually ignores the estimated $50,000 in engineer time, bonuses, and ramp-up costs involved in hiring.
  • The author notes that no one talks about how a highly paid employee is now focused entirely on infrastructure work that does not directly advance the original mission.
  • They have seen leaders complain about the cost of support, team events, and software services, but not about engineers shifting from feature delivery to infrastructure maintenance.
  • A year later, GoCo is thriving and plans a GoCoCon, while the new operations team is praised for metrics like 99.999% uptime.
  • Despite these successes, leadership still does not ask what a platform like Heroku would cost compared with the full expense of running an internal operations team.
  • The author estimates that a team of three to four operations engineers can cost about half a million dollars per year when including salary, benefits, stock, hiring and bonuses, communication overhead, and office costs.
  • They call this $500,000 figure a conservative guess and argue managers should reevaluate Platform-as-a-Service options.
  • The comparison between PaaS and low-cost virtual machines should include the cost of the engineers required and the impact on the team when attention drifts away from the mission.