将“赋能他人”作为工作的目标。

Like every software engineer on the planet, I have thoughts about how AI is reshaping our jobs. At Hidden Door we’ve been using language models large and small, alongside a bevy of NLP techniques, since our inception (circa GPT-2, if that’s how you want to think about time). This year, there has been a step change in our adoption of them outside of the product, in doing our work. We have, like, a bunch of bots, and naturally we’ve had a period of adjusting to our new roommates.

I am Hidden Door’s pet code-quality wonk, and practically allergic to the writing tics of LLMs. How have I, a craft-minded engineer, made my peace? By prioritizing enablement. I see my colleagues who do not wear an engineering hat being able to do things that they would previously have had to ask an engineer for, and it makes me happy.

Those of us with engineering titles are often custodians of assorted keys to the product castle. We’ve opened some of the castle doors. An analytics query here, a copy edit in the guts of the game experience there. Tools, rightly placed, can greatly enhance people’s agency to contribute to building the thing. I am not saying I’m glad I have less work now. I have plenty to do. Folks are asking more questions of our data (for instance), because a small hurdle has been removed.

My natural inclination as a software maker is to turn everything into tools. And tools are great! Our bots, tools themselves, are natural language interfaces to other sets of tools. And do they ever make some things easier! I’ve always been a bottom-up kind of thinker, I love to understand things! But do I really want to maintain knowledge of the command line APIs for a whole host of third party telemetry and application monitoring services? Reader, I do not. One of our bots—we call it Tobi¹ —can do that for me, and does not blink at correlating information across three external services and our internal databases. It is rad.

Tobi has skills for writing good Linear tickets, triaging which can be sent on to additional automated dev, gathering information to help debug something or respond to an incident, and knowing when to bring things to our human attention for decisions. Because it’s in our chat, everyone on the team can ask natural language questions of our systems, bounce around an idea for a feature (and get a quick idea of how easy or hard it might be to implement), and gather information to help support our players or make better social posts. Being in-chat also makes it collaborative. We can tap each other in as needed.

How we do it

Using technology to enable each other is not restricted to putting a bot in team chat. Examples of enabling things at Hidden Door:

• Agent skills for guiding a human through adding art to our game, or creating a new earnable achievement, and systems designed to make those things easy.
• A task runner that makes it easy to run the right tests, lints, typechecks and co, the same way for all.
• Evaluation harnesses for content authoring, which can give a sense of how some content will work in the system (always alongside plenty of experiential testing).
• A small minigame framework so that non-coders can prompt their way to minigames, and the ability to easily playtest and validate them.
• An automated triage pipeline that gathers information together into a ticket, so an engineer can make an easier call.
• Countless code-architecture decisions that try to make the right thing obvious and the wrong thing hard.
• And, yes, the bots.

Being enabling as an engineer doesn’t even have to mean building things at all. You can foster a culture of enablement. Just tell a colleague to go ahead and try something that maybe they aren’t sure they can do, and that you will work on it together. Give them a tour of that metaphorical castle. I’m not arguing that your marketing guru should own the next database migration, but I do think that building software is becoming ever more holistic. This trend did not start when LLMs entered the zeitgeist, but they have accelerated it.

It’s a little uncool to talk about code these days, but as the bots are bringing others into the realm of source code, we code-lovers might extend outwards the behaviours that make one a good citizen of a codebase. Always try to be designing and refining such that the system gets easier to change, the right thing gets easier to do, mistakes are harder to make, invariants harder to break. Leave things better than you found them. Such principles apply as much to work aided by bots as at any other time.

I don’t think the software engineer is yet endangered, and I do think it is good to always be empowering the people around you. Doing that in a way that does not overly burden them with responsibilities is the engineering. “You can query analytics now but if you ask for a wrong thing you bring down prod” is no power at all. If you make enabling others the goal of your work, you aren’t writing yourself out of a job, you’re giving your co-workers cool new powers!