AI Native Engineering.

AI adoption in companies is no longer a decision to make: it has already happened, with or without governance. Developers use it every day, the volume of code grows at an unprecedented pace, and the real problems have moved to an organizational level that no vendor talks about.

• How do you measure the return on investment when apparent productivity explodes, but reviewing all that code relies on human cognitive capacity that does not scale at the same pace?
• How do you know whether teams are heading in the right direction, building skills or only delegating?
• How do you keep adoption sustainable, caught between the growing cost of tokens on one side and the mental load of those who validate, integrate and maintain on the other?

And above all: if writing code is no longer the bottleneck, what is your team’s new goal?

In this talk we bring the perspective of those who lead technical teams through this transition: how to help a developer move from execution to orchestration, a role that delegates implementation to a set of agents and focuses on defining features, validation criteria and real impact, without cognitive overload.

Which indicators tell real growth from illusory productivity. Which concrete tools to put in place, from spending boundaries to rethought review practices, from team rituals to growth paths to rewrite.

Because the real ROI of AI is not measured in lines of generated code, but in the organization’s ability to scale without burning margins, quality and people.

Agentic coding promises productivity, but it has a hidden cost: tokens. A naive setup burns budget and context in a few hours and also lowers the quality of the answers.

Our setup is the result of months of tuning, measurement and instructive mistakes. We share:

• which tools we chose and why;
• how we structure project rules, memory and context;
• which techniques we use to cut consumption: from context management to delegating to cheaper models, up to targeted skills and automations.

A practical path from occasional use of agents to a sustainable daily practice.

What if an agent could take a frontend task from the issue to ready-to-review code, leaving the developer only the review? This is the flow we are building at SparkFabrik, tested on real projects.

Davide and Enrico take turns on design, tools and implementation:

• deterministic extraction of Figma data;
• small skills the whole team can reuse;
• automatic mapping of design tokens;
• component generation for Drupal, React, Angular and Vue.

What already works, what we are still stabilizing and how work changes when manual file edits become the exception.