How platform teams get stuff done

"Platform" > Internal Developer Platform

There's a lot of buzz at the moment around Platform Engineering, primarily focused on Internal Developer Platforms (IDPs). I want to make it clear that the discussion of "platforms" here is significantly broader; it encompasses other internal products such as a data platform, a front-end design system, or an experimentation platform.

In fact, while we will be primarily concerned with technical platforms, a lot of the ideas presented here also apply to internal products that provide shared business capabilities - a money movement service at a fintech company, or a product catalog service at an e-comm company. The unifying characteristic is that platforms are internal products used by other teams within an organization. Thus, platform teams are building products whose customers are other teams within their company.

Examples of migration work

What types of changes are we talking about? One relatively simple migration would be a version upgrade- upgrading a shared component library, or upgrading a service's underlying language runtime.

A common, larger migration would be replacing direct integration of a 3rd party system with some internal wrapper - for example, moving logging, analytics, or observability instrumentation over to using a shared internal library maintained by a platform team, or replacing direct integration with a payment processor with integration via an internal gateway service of some kind.

Another type of migration might be replacing an existing integration into a deprecated internal service with an integration into it's replacement - perhaps moving from an old User service to a new Account Profile service, or migrating usage of a credit-puller and credit-reporting service to a new consolidated credit-agency-gateway service.

A final example would be an infrastructure-level re-platforming - dockerizing a service owned by a product team, introducing a service mesh, switching a service's database from MySQL to Postgres, that sort of thing.

Note that with platform migrations the product team is often not especially motivated to make these changes. Sometimes they are, if the new platform is going to provide some particularly exciting new capabilities, but often they are being asked to make this shift as part of a broader architectural initiative without actually getting a huge amount of value themselves.

Collaboration Patterns

Let’s look at what cross-team collaboration patterns would work for platform migration work.

Farm out the work

The platform team could File a Ticket in the product teams’ backlogs, asking them to make the required changes themselves.

This approach has some advantages. It’s scalable - the implementation work can be farmed out to all the product teams whose codebases need work. It’s also trackable and easy to manage - often the ticket filing can be done by a program manager or other project management type.

However, there are also some drawbacks. It’s really slow - there will be long lead times before some product teams get around to even starting the work. Also, it requires prioritization arm-wrestling - the teams being asked to do this work often don’t receive tangible benefits, so it’s natural that they’re included to de-prioritize this work over other tasks that are more urgent or impactful.

Platform team does the work

The platform team might opt to make changes to the product team's codebases themselves, using three similar but distinct patterns - Tour of Duty, Trusted Outsider, or Internal Open Source.

With Tour of Duty, an engineer from the platform team would "embed" with the product team and do the work from there.

With Trusted Outsider and Internal Open Source the product team would accept pull requests to their codebase from an engineer in the platform team.

Just as with taking the file-a-ticket path, having the platform team do the work comes with some pros and cons.

On the plus side, this approach often reduces the lead time to get changes made, because the team that needs the work to be done (the platform team) is also the one doing the work. Aligned incentives mean that the platform team is much more likely to prioritize their work than the product team which owns the codebase would.

On the negative side, having the platform team do the migration work themselves only works if the product team can support it. They either need to be comfortable with a platform engineer joining their team for a while, or they need to have already spent enough time with a platform engineer that they trust them to make changes to their codebase independently, or they need to have made the significant investment required to support an internal open-source approach.

Another negative is that this do-it-yourself strategy is not scalable. There will always be less engineering capacity on the platform team compared to the product delivery teams, and not delegating engineering work out to the product teams leaves all that capacity on the table.

Really, it's a bit more complicated

In reality, what often happens is a combination of these approaches. A platform team tasked with a migration might have a program manager file tickets with 15 product delivery teams and then spend some period of time cajoling them to do the work. After a while, some teams will have done the work themselves but there will be stragglers who are particularly busy with other things, or just particularly disinclined to take on the migration work. The platform team will then roll up their sleeves and use some of the other, less scalable approaches and make the changes themselves.

Collaboration patterns

A worthy goal for many platform teams is to build a fully self-service platform - something like Stripe or Auth0 that’s so well-documented and easy to use that product engineers can use the platform without needing any direct support or collaboration with the platform team.

In reality, most internal platforms aren't quite there, especially early on. Product engineers getting started with an internal platform will often run into poor documentation, obtuse error messages, and confusing bugs. Often these product teams will throw up their hands and ask the platform team to pitch in to help them get started using the features of an internal platform.[1]

When a platform consumer is asking the platform owner for hands-on support we are back to cross-team collaboration, and once again different patterns come into play.

Professional services

Sometimes a product team might ask the platform team to write the platform consumption code for them. This might be because the product team is struggling to figure out how to use the platform. Or it could be because this approach would require less effort from the product team. Sometimes it's just a misunderstanding where the product team doesn't think they're supposed to do the work themselves - this can happen when shifting into a devops model where product teams are self-servicing their infra needs, for example.

In this scenario the platform team sort of becomes a little professional services group within the engineering org, integrating their product into their customer’s systems on their behalf.

This professional services model uses a combination of collaboration patterns. Firstly, a product team will typically File a Ticket requesting the platform team's services. This is the same pattern we looked at earlier for Platform Migration work, but inverted - in this situation it's the product team filing a ticket w. the platform team, asking for their help. The platform team can then actually perform the work using either the Trusted Outsider or Internal Open Source patterns.

A common example of this collaboration model is when a product team needs some infrastructure changes. They want to spin up a new service, register a new external endpoint with an API gateway, or update some configuration values, so they file a ticket with a platform team asking them to make the appropriate changes.

This pattern is commonly seen in the infra space, because it perpetuates an existing habit - before self-service infra, filing a ticket would have been the standard mechanism for a product team to get an infrastructure change made.

White-glove onboarding

For a platform that's in its early stages and lacking in good documentation, a platform team might opt to onboarding new product teams using a "white glove" approach, working side-by-side with these early adopters to get them started. This can help kickstart the adoption of a new platform by making it less onerous for the product teams who go first. It can also give a platform team really valuable insights into how their first customers actually use the platform's features.

This white-glove model is typically achieved using the Tour of Duty collaboration pattern - one or more platform engineers will spend some time embedded into the consuming team, doing the required platform integration work from within that team.

Community of practice

As a platform matures and becomes easier to use it a platform team can step away from doing hands-on integration work and move into a more consultative role.

This consultative mode includes things like hosting "office hours" where a consuming team can show up and ask questions, or having a platform representative providing focused advice and guidance to a consuming team's planning and design sessions. In Team Topologies parlance this would be the platform team operating in a facilitating interaction mode.

For large, rich platforms there is eventually a move towards a peer support model, where a platform team invests time to stand up a Community of Practice for users of their platform. This can involve things like a community slack channel or mailing list, a regular community of practice meeting to seek help and showcase interesting ideas, perhaps even an annual practitioners offsite.

Hands-on doesn't scale

We can see that the level of hands-on support that a platform team needs to provide to consumers can vary a lot depending on how mature a platform's Developer Experience is - how well it's documented, how easy it is to integrate and operate against.

In the early days of a platform, it makes sense for platform consumption to require a lot of energy from the platform team itself. The developer experience is still a little rocky, platform capabilities are perhaps still being built out, and consuming teams are perhaps a little skeptical to invest their own time as guinea pigs. What's more, working side-by-side with product teams is a great way for a platform team to understand their customers and what they need!

However hands-on support doesn't scale, and if broad platform adoption is the goal then a platform team must invest in the developer experience of their platform to avoid drowning in implementation work.

It's also important to clearly communicate to platform users what support model they should expect. A product team that has received white-glove support in the early days of platform adoption will look forward to enjoying that experience again in the future unless informed otherwise!

File a ticket

The product team could File a Ticket with the platform team, asking them to make the required changes to their platform. This tends to be a very frustrating approach. Often a product team only realizes that the platform is missing something at the moment that they need it, and the turnaround time for getting the platform team to prioritize and perform the work can be way too long - platform teams are typically overloaded with inbound requests. This leads to the platform team becoming a bottleneck and blocking the product delivery team’s progress.

Move engineers to the work

With sufficient warning, teams can plan to fill a gap in platform capabilities by temporarily re-assigning engineers to work on the required platform enhancements. Product engineers could do a Tour of Duty on the platform team, or alternatively a platform engineer could join the product team for a while as an Embedded Expert.

Moving engineers between teams will inevitably lead to a short-term impact on productivity, but having an embedded engineer can increase efficiency in the long run by reducing the amount of cross-team communication that's needed between the product and the platform teams. The embedded engineer acts as an ambassador, smoothing the communication pathways and reducing the games of telephone.

This equation of fixed upfront costs and ongoing benefits means that re-assigning engineers is an option best reserved for larger platform improvements - moving an engineer to another team for a couple of weeks would be more disruptive than helpful.

These types of temporary assignments also require a relatively mature management structure to avoid embedded engineers feeling isolated. With an Embedded Expert - a platform engineer re-assigned to a product team - there’s also a risk that they become a general “extra hand” who’s just doing platform consumption work, rather than actively working on the improvements to the platform that the product team need.

Work on the platform from afar

If a platform team has embraced an Internal Open Source approach then a product team has the option of directly implementing the required platform changes themselves. The platform team's role would be mostly consultative, providing design recommendations and reviewing the product team's PRs. After a few PRs, a product engineer might even gain enough trust from the platform team to be granted the commit bit and become a Trusted Outsider.

Many platform teams aspire to get to this situation - wouldn't it be great if your customers were empowered to implement their own improvements, and save you from having to do the work! However, the reality of internal open-source is similar to open-source in general - it takes a surprising amount of investment to support external contributions, and the large majority of consumers don't become meaningful contributors.

Platform teams should be careful to not open up their codebase to external contributions without making some thoughtful investments to support those contributions. There can be deep frustration all around if a platform team proudly proclaim in an all-hands that their codebase is a shared resource, but then find themselves repeatedly telling contributors from other teams "no, no, not like THAT!".

be collaborative, but only for a bit

In Team Topologies, they point out that the best way to design good boundaries between two teams is to initially work together in a focused, very collaborative mode - think of patterns like Embedded Expert and Tour of Duty. This period can be used to explore where the best boundaries and interfaces to create between systems, and between teams (Conway's Law tells us that these two are inextricably entwined). However, the authors of Team Topologies also warn that it's important to not stay in this collaborative mode for too long. A platform team should be working hard to define their interfaces, looking to move quickly to an "as-a-service" mode, using patterns like File a Ticket and Internal Open Source. As we discussed in the Platform Consumption section, the more collaborative interaction models simply won't scale as far as the platform team is concerned. Additionally, collaborative modes impose a much greater cognitive load on the consuming teams - moving to more hands-off interaction styles allows product delivery teams to spend more of their time focused on their own outcomes. In fact, Team Topologies considers this reduction of cognitive load as the defining purpose of a platform team - a framing which I very much agree with.

Navigating this shift from highly collaborative to as-a-service is, in my opinion, one of the biggest challenges that a young platform team faces. Your customers become comfortable with the high-touch experience. Building great documentation is hard. Saying no is hard.

Platform teams operating in a collaborative mode should be keeping a weather eye for scaling challenges. As the need for a shift towards a more scalable, hands-off approach appears on the horizon the platform team should begin signaling this shift to their customers. An early warning as to how the interaction model will change - and why - gives product teams a chance to prepare and to start shifting their mental model of the platform towards something that's more self-sufficient.

The transition can be painful, but vacillating makes it worse. A product delivery team will appreciate clearly communicated rules of engagement around how their platform providers will support them. Additionally, removing the crutch of hands-on collaboration provides a strong motivation to improve self-service interfaces, documentation, and so on. Conway's Law is in effect here - redefining how teams integrate will put pressure on how the team's systems integrate.