Developers build the best tools for developers — and are now defanging the AI menace

Software engineers absorbed more change in three years than in the previous thirty. That reality hit hard at AI Engineer Melbourne, where speakers walked through the emotional and practical terrain of building alongside machines that increasingly do what they trained to do.

Fear and grief are natural reactions to tools that automate your work. The next reactions — acceptance and innovation — prove more useful.

AJ Fisher presented on diffusion models, the same class of system that generates images. Fisher's approach: use a cheaper, faster model and iterate until the result holds. The strategy, sometimes called the "Ralph Wiggum loop," delivers comparable output to pricier frontier models at one-half to one-tenth the cost.

Google released its DiffusionGemma model days after Fisher's talk, giving engineers access to text generation at scale without the token bill that burns through budgets.

Organizations have already shifted from flat-rate subscriptions to metered token pricing. Several conference speakers addressed managing those costs as AI tooling moves from experiment to infrastructure.

Annie Vella, author of "The Software Engineering Identity Crisis," shared research on how AI tooling triggers grief in engineers. The field has split into all-in and never-ever camps, with a broad middle testing the water.

The divide traces to two work styles. Engineers who value the journey of learning feel short-circuited when AI skips steps. Engineers who chase outcomes see AI as leverage.

Vella proposed sensitivity, listening, and openness to change on both sides.

Jeremy Howard, Kaggle and fast.ai alum, pushed back against the drift toward passive consumption of machine output. Howard demonstrated SolveIT, a beta tool combining Python notebooks, Mathematica, Wikipedia, and chatbot functionality into a single environment designed for active engagement with knowledge rather than mindless absorption.

Howard's argument: critical thinking is the skill that keeps engineers relevant, not the ability to prompt a model.

Daniel Rodgers-Pryor closed with a working demonstration of what he called "Fully Automated Luxury Gay Space Engineering." His entire CI/CD pipeline feeds metrics, logs, and user feedback into AI agents that identify issues, locate root causes, fix problems, run tests, and deploy updates.

The result is a self-healing system that improves as pressure increases. More users generate more data. More data improves the system.

Rodgers-Pryor compared the loop to a production line worker sampling bonbons for quality and returning them to the stream. The question he posed to the audience: how do you make those feedback loops shorter and tighter?

The conference exposed a tension at the center of software engineering right now. Engineers have every right to feel aggrieved by the pace of change. But the speakers at AI Engineer Melbourne — Fisher, Vella, Howard, and Rodgers-Pryor — framed adoption not as capitulation but as exploration.

New tools carry risk. They also carry the kind of opportunity that comes once in a career.

The author attended AI Engineer Melbourne as a guest of the conference.