Engineering strategy for scaling a subscription platform
This guide explains how a subscription-infrastructure engineering organization structures its work as it grows: small autonomous teams with clear technical leads, aggressive use of AI tools with full ownership of what ships, transparent incident handling, and a global remote hiring model. It is written for engineering leaders and senior engineers who need to adapt these decisions to their own teams.
Mission and scope
The engineering team's stated mission is to make great software that helps developers make more money. Internally, the success of every other team is downstream of what engineering builds. Externally, the platform is a critical component for tens of thousands of apps, affecting monetization, analytics, growth, and even app store ratings. That trust is treated as something that can be lost in an instant.
The scope of work covers three areas:
- Build — subscription infrastructure and data accuracy. A single minute of downtime can mean nearly a million failed API requests, and a buggy SDK can crash tens of thousands of apps.
- Analyze — business analytics. Inaccurate data can lead customers to overspend on ads or send incorrect reports to investors.
- Grow — growth tools such as experiments, paywalls, and targeting. Because the business model is tied to customer revenue, helping developers earn more is a shared win.
The organization reports serving over 100K apps, maintaining 10 SDKs, processing $14B+ annually, and handling 3B+ API requests per day. These figures are a dated reference from the published strategy and should be checked against current conditions.
Team structure
As of June 2026, the engineering organization included over 100 fully remote team members across 21 countries and 11 time zones, organized into 19 cross-functional teams covering customer-facing, product, data, and infrastructure work. Two technical support teams exist: Development Support Engineering for self-serve customers, and Technical Account Management for implementations and enterprise customers.
A small group of senior individual contributors, OCTO (Office of the CTO), reports to the CTO. OCTO focuses on bootstrapping new bets, unblocking critical projects, and carrying technical and cultural standards into ambiguous work. OCTO is not a replacement for teams; it starts or rescues important work until durable ownership exists.
Decision checklist for splitting a team:
- Is ownership, speed, or decision-making unclear for a product surface area? If yes, prefer a smaller autonomous team with a clear Engineering DRI.
- Can an engineering manager oversee multiple small teams while keeping performance management separate from the DRI role?
- Does the work fit an existing team, or does it need startup-level speed without overloading current teams? If the latter, consider an OCTO-style unit.
- Review team creation at least twice a year.
Guiding principles: on building
The strategy sets out principles in three categories: building, incidents, and growth.
On building:
- "Almost done" is worse than "not started." Shipping velocity is the standard the whole company looks to. Reduce scope relentlessly, deliver the minimum viable product, then analyze, optimize, and iterate.
- Small teams move faster. Large teams create coordination overhead, consensus loops, and unclear ownership. Small teams with a clear Engineering DRI can decide closer to the work and show progress weekly. AI increases the leverage of small groups, but managers and product leaders must keep the product cohesive.
- Fires over features. When something breaks or affects customers — outages, crashes, data inaccuracies, regressions — it takes priority over unshipped work. Context switching is hard, but this is how the platform keeps improving.
- No change is innocuous. A small deployment can affect the whole system, and a minor metric change can go unnoticed. Seemingly small changes often get quicker, less thorough reviews. Plan, monitor, and treat testing as mandatory.
- Do the dirty work. App stores change constantly and carry bugs and inconsistencies. The job is to abstract those issues so developers do not have to write store-specific branches themselves.
- Use AI, but own the output. Everyone should actively use AI tools and learn to get leverage from them. AI does not change ownership: even if an agent writes the code, the engineer owns what ships. It is acceptable to open pull requests you could not have written without AI, as long as you fully understand the output — read the code, ask AI to explain or review it, and only then ask a teammate to look. Otherwise maintainers, especially in large shared codebases like frontend, get overwhelmed.
- Be pragmatic. Prefer proven, stable technologies. No full rewrites. Introduce new technology only when the current stack has issues, and then build a strong case, outline a migration path, and give team members and leadership a chance to review.
Worked example — fires over features: Suppose a team is mid-way through a new paywall feature and a regression is reported in subscription data accuracy. Under this principle, the regression takes priority over the unshipped feature. The team pauses feature work, investigates, and only returns to the feature once the customer-affecting issue is resolved. The 3B+ daily API request figure shows why: a small data inaccuracy can propagate to many customers' reports.
Guiding principles: on incidents
On incidents:
- Escalate early. Aim to detect problems before customers do. If a customer or teammate notices something unusual, investigate immediately and escalate per the incident response guidelines. Create an incident and centralize communications in the designated channel. A false alarm is better than ignoring a potentially significant issue.
- Do not hide the nasty stuff. Hiding a problem never works: people eventually notice, and the damage compounds. Being open is how trust gets built. For anything customers can see, post a public incident on the status page right away and keep that post current as the situation evolves.
- Analyze and automate. Mistakes happen and there is no blame in making them; there is blame in not taking responsibility and learning. Once the root cause is identified, documenting it is not enough — implement automations so others do not have to think about it and will not repeat the mistake.
Incident flow map:
The organization continues to use Rootly for incident management and maintains the practice of blameless post-mortems.
Guiding principles: on growth
On growth:
- Build a winning team. The organization chose a global remote model with the same pay everywhere. That only works if extraordinary people are brought onto the team, so hiring, promotion, and performance management must be done exceptionally well. Every seat counts.
- Use OCTO for ambiguous, high-leverage work. New bets, critical technical problems, and ambiguous early-stage initiatives often need startup-level speed without diluting culture or overloading current teams. OCTO is a small group of trusted senior ICs who bootstrap new bets, unblock important projects, and provide a signal on technical and cultural health. It is not a replacement for teams.
- Learn and educate about the space. New hires are not expected to be domain experts on day one, but everyone is expected to keep deepening their understanding of the space and customers' challenges — by engaging with customers, listening to the Sub Club podcast, attending conferences, replying to support tickets, and using the product.
- Support and infrastructure cost should scale sublinearly. Processes and human and computer costs should improve with economies of scale. Every support ticket category is treated as a sign of a product or documentation failure to be improved.
Coherent actions: turning strategy into reality
The strategy is backed by a set of continuing and one-off actions:
- Prefer smaller autonomous teams with clear Engineering DRIs, especially where ownership, speed, or decision-making are unclear.
- Engineering managers oversee multiple small teams when appropriate, with performance management kept separate from the DRI role.
- Use OCTO to bootstrap new bets, unblock critical projects, and carry standards into ambiguous work until a durable owner exists.
- Monitor and support the creation of new teams or sub-teams, ideally at least twice a year.
- Continue using Rootly for incident management and keep writing blameless post-mortems.
- Add new stores or payment providers to reduce complexity for customers; the number of potential additions is finite, and a few are expected each year.
- Follow a standard hiring interview loop so every candidate goes through the same process, even as the loop evolves.
- For customer-facing roles, value experience in mobile apps, in-app subscriptions, or familiarity with the platform, since it speeds onboarding.
- Rework the uptime definition so it also captures signals that end users never see but that customers depend on — for example, developer-facing system behavior that erodes confidence even when app users notice nothing. That broader measure becomes the basis for tracking and setting an ambitious long-term reliability target.
What's next
The focus remains on building reliable, scalable, developer-loved tools, refining processes, learning from customers, and pushing what the platform can do. The organization states it is always looking for talented people to join the team.
| Stage | Action | Owner signal |
|---|---|---|
| Detect | Monitor; act on customer or teammate reports | Any engineer |
| Escalate | Create incident; centralize comms in designated channel | Incident responder |
| Communicate | Open public status page incident if customer-facing; update regularly | Incident responder |
| Learn | Write blameless post-mortem; identify root cause | Team |
| Automate | Implement automation to prevent recurrence | Team |
Frequently asked questions
What does 'almost done is worse than not started' mean in practice?
It means a feature that is 90% complete but not shipped delivers no validated value. The principle pushes teams to reduce scope, ship a minimum viable product, and iterate based on real user feedback rather than polishing unshipped work.
How does AI-assisted development affect code ownership?
It does not change ownership. Even if an AI agent writes the code, the engineer owns what ships. It is acceptable to open pull requests you could not have written without AI, as long as you fully understand the output — read the code, ask AI to explain or review it, and only then ask a teammate to review.
When should a new team be created?
When ownership, speed, or decision-making are unclear for a product surface area. The preference is for smaller autonomous teams with a clear Engineering DRI, and team creation is reviewed at least twice a year.
What is OCTO and when is it used?
OCTO (Office of the CTO) is a small group of senior individual contributors reporting to the CTO. It is a temporary vehicle for launching new bets, clearing blockers on important projects, and instilling technical and cultural expectations in work that lacks a settled owner — it hands off once a durable team or owner takes over. It is not a replacement for teams.
How are incidents handled?
Escalate early, centralize communications in a designated channel, open a public status page incident if customer-facing, write a blameless post-mortem, and implement automation to prevent recurrence. A false alarm is preferred over ignoring a potentially significant issue.