How I run things

Much of leading technical work is understanding what people are trying to accomplish, what they already know, and what is getting in their way. The architecture matters, but so do the responsibilities and expectations around it. I want the people doing the work to understand the decisions that affect them and to have room to challenge those decisions.

The platform should learn the people

A data engineer who understands the business rules and can express them in SQL already has valuable expertise. At Ticketmaster, the Databricks tooling let engineers use that expertise without first having to learn Spark configuration, scheduling, and recovery. At ArtsEd West, the same consideration leads to a different interface: my cofounders can propose and review website changes through Google Sheets because that is a tool they already use.

Guardrails, not gates

I want teams to have enough ownership to make decisions about their work, with clear boundaries around what affects other people. At Xero, shared modeling tools let product teams develop their own models while the platform managed access, schedules, and dependencies. The point of those controls was to make that independence practical. Coding agents need boundaries too, which is why their development access stops short of production credentials and promotion.

The work should outlast me

Leaving a roadmap is only part of a handover; someone needs the authority and capacity to keep delivering it. At Xero I developed a successor and stayed through the transition, with protected teams already responsible for the replacement. They carried that work through after I left and delivered the real-time path in 2025. Working myself out of the job means the team can continue without depending on my presence.

Evidence over ego

At Ticketmaster I wrote a Kafka Streams implementation while a principal engineer and his team built the competing Snowflake path. Comparing them meant assessing my own work against theirs, including how difficult each would be for other engineers to maintain. Snowflake was easier to scale and change, so I retired my implementation and kept the system the organization could more readily support.

Trust takes repeated work

At Xero, Revenue Operations and product analysts had built their own tools because they needed to get work done. Asking them to use a shared platform meant understanding what those tools did for them and making their experience part of its development. We combined their tooling with the engineering framework, and it took about 18 months for both groups to be using and supporting it. Their confidence in the platform had to develop through that work.

Technical judgment should help us move

I need enough technical understanding to discuss a difficult choice with the engineers making it, while leaving them room to do their work. At Radix, reading the implementation let me question the coupling between the new frontend and the existing financial logic. The resulting design preserved that logic initially and gave the engineers boundaries through which they could change individual capabilities, rather than requiring them to rewrite everything before making progress.

People should know where they stand

Expectations affect people's careers, so a roles framework needs to be useful to the person doing the job as well as to their manager. At Radix I defined responsibilities across engineering, analytics, and science and simplified titles and pay bands. I also worked with managers on a performance framework at Xero. In that work, I want people to understand what is expected of them and how they can develop, with room to discuss where the expectations do not fit the work.

When I ask someone to take ownership, I need to be clear about which decisions they can make, what support they can expect, and which constraints remain mine to resolve. A manager cannot reasonably promise autonomy and then require every meaningful choice to return for approval. Equally, leaving someone alone with an unclear objective is not a useful form of trust.

Performance expectations need the same care. I want people to hear about a concern while there is time to understand it and act, with specific examples and room to explain what I may have missed. The framework should improve that conversation, rather than become a reason for the manager to avoid having it.

Retire the old thing by making the new one better

People can have good reasons for staying on an old system, particularly when they understand its limitations and depend on it every day. At Xero, the replacement offered shared tooling and support, while dedicated teams had capacity to move the work across. We reduced legacy processing and support as that happened. A migration plan has to account for the people making the move and the work they still need to finish while it is underway.

The newest user is an AI

I have about seventy agents configured on a server in the closet, working within development environments. They can investigate a problem and propose changes, but the people who use the software still need to be able to judge whether those changes help. That is why the next part of the repair workflow is a way for the person who reported a problem to try the proposed fix and ask for revisions.