Of the work my team delivered last year, I can think of almost nothing that involved building a capability that did not already exist somewhere we could buy. What we did, over and over, was make two or three things that were never designed for each other behave as one coherent process: an identity provider, a finance package, a sector-specific application from a vendor with eleven customers, and a spreadsheet that turned out to be load-bearing. The value we added was almost entirely in the joins.
I do not think that is a phase we are passing through. It is the shape the industry has settled into, and the shape it is getting further into. Producing a plausible application is now cheap and still getting cheaper. What has not got cheaper by any measurable amount is making something fit a real organisation's existing obligations, data, access model, audit requirements and twenty years of accumulated exceptions. The scarce thing was never code. It is now very obviously not code.
The consequence I am planning around is that the integration surface grows faster than the number of systems. More systems, built faster, by more people, each needing to exchange data with the others, each with its own notion of what a customer is and when a record is authoritative. Every one of those joins is a place where meaning has to be reconciled, and reconciling meaning is a judgement call that requires knowing the business, not just the protocol.
So the skills I am deliberately keeping sharp are unfashionable ones. Reading somebody else's API documentation critically and working out what it does not say. Designing contracts between systems that survive one side changing. Identity and authorisation across boundaries, because that is where most of the real risk now lives. Knowing enough about how the business actually operates to tell which of two conflicting records is true. Being able to explain a trade-off to a manager who will decide based on what I say.
None of that is glamorous and none of it demos well. But I notice that every genuinely hard problem that lands on my desk is a problem about the space between systems, and I see nothing coming that reduces the number of spaces.
– Serguey Shinder
Top comments (0)