# Shipping fast doesn't equal value: Judgment, understanding, and humanity are the new competitive edge > Source: > Published: 2026-08-14 18:54:50+00:00 # Shipping fast doesn't equal value: Judgment, understanding, and humanity are the new competitive edge Code & Theory, Braze, and Sanity leaders on AI productivity, hiring, and winning over skeptics. Kelsey Bernius Staff Enterprise Product Marketing Manager at Sanity Published The proliferation of coding has enabled anyone, regardless of technical acumen or background, to vibe code a new feature and lob it over to the engineering team to ship. But thousands of dollars worth of tokens later, how often does it actually get shipped and provide value? The answer for many is an uncomfortable one: not yet. AI can do many things for us, but human teams face the same challenges and questions: How do you ensure you ship the right features, hire the right talent, and convince stakeholders that their workflows require a complete paradigm shift? Three engineering leaders — Dave DiCamillo (CTO, Code & Theory), David Willis (VP of Data and Growth, Braze), and David Annez (SVP of Engineering, Sanity) — recently compared notes on these topics at [Sanity Context](https://www.youtube.com/watch?v=4IFy_61N8q4), our annual gathering of engineering and content leaders. Everyone agreed that judgment is becoming more valuable than raw output, the distinctions between engineers, designers, and product managers are dissolving, and the hardest problems left aren't technical anymore but rather convincing the entire company to adopt new processes and workflows. Read on for more insights across these themes. Shipping fast and shipping right are different skills DiCamillo explained that the easiest thing to build with AI right now is something that *looks* finished. But his team spends a significant amount of time educating clients on the difference between vibe-coding a prototype and building something secure, performant, and reliable at real scale. “The last two years have handed serious engineering tools to a lot of people who never had them before — which is genuinely good, since it's brought creatives and strategists much closer to the code itself, arriving with actual pull requests instead of just a requirements doc,” said DiCamillo. “But it also means more requests now arrive looking production-ready when they aren't, and someone still has to do the unglamorous work of making them actually hold up.” That same gap is reshaping how Code & Theory hires. Interviews used to assess what a candidate had already built by looking at their repo and previous projects. Now, candidates spend the first hour of an interview using AI tools against a live brief, then the last thirty minutes talking through flaws that they caught and where the AI fell down. Raw technical output is not being evaluated anymore. A candidate's rationale is far more important. Everyone manages teams of agents Annez noted that independent contributors are increasingly required to think like leaders because, in practice, they're managing a team of agents. “The job is less about writing every line and more about a kind of abstraction and defining a problem precisely enough that an agent can actually execute against it well,” said Annez. Willis explained the same blur from the product side. “Engineers increasingly need to understand the what and the why themselves, or teams risk stitching together a Frankenstein's monster of a product, one that technically works but was never held together by a single clear goal,” said Willis. But all of this work requires resources. Annez identified the real cost as cognitive debt. When a team can ship ten times more code in the same amount of time, understanding how any of it actually works becomes much harder to keep pace with. Everyone still has to test it, and testing still takes time. The speed of generation doesn't buy you a shortcut on verification. The ROI question hasn't changed Inevitably all AI conversations lead back to ROI. Willis pointed out that measuring return on AI investment isn't actually a new problem. The real test is still whether something drives customer acquisition, not how many features engineering shipped last sprint. “What *is* new is the cost structure underneath that question — tokens spent per pull request, the real dollar cost of giving every developer a meaningful AI budget, and the discipline required to tie that spend back to actual product outcomes instead of just velocity for its own sake,” said Willis. Structured content is….everything All three leaders agreed that as AI systems take on more real work, what they're reading and acting on matters more than anything. DiCamillo argued that structuring content as data, rather than as a loose pile of assorted assets, is exactly the foundation AI needs to work reliably, and that the further back a clean system of record goes, the more a business can actually trust it. Willis recalled the first time his team connected Slack and Google Drive into a single AI system, the improvement in what they could actually find (data fidelity, as he put it) was immediate. “The real work then becomes getting everything into one governed system, so content can be created once and applied consistently everywhere it needs to live,” says Willis. Annez tied the concept together at the infrastructure level and said the ability to pull content in from anywhere and structure it is becoming a workflow primitive. “The moment an organization can ingest content into a real single source of truth is the moment AI and automation start to get valuable, instead of just automating one narrow task or chat at a time,” says Annez. The hardest problem left is adoption, not technology Annez described how AI is changing the entire publishing process within companies. Historically, a person opened a document and decided what needed to change. In agentic content workflows, an agent surfaces what should change and presents that to the publisher. The human's role shifts toward review rather than authorship (pull vs. push relationship). He said the ultimate goal, and what he is building at Sanity, is for a content platform to become a space of confident publishing or somewhere a person can trust what's being surfaced enough to sign off quickly, without redoing the underlying work themselves. With these changes, comes hesitation from some. DiCamillo says there’s still a lot of diehards who just don’t want to give up their publishing process because they’ve been doing it for the past 20 years and it already works. “Just because we can build something doesn’t mean somebody is going to use it.” said DiCamillo. “We spend an inordinate amount of time educating our clients to understand the workflow changes that are going to be coming.” Change management, not capability, is where most of the real friction now lives. It’s hard to stomach but everyone agrees it is a necessary journey to go on. “The reality is that you win or you lose and the old schoolers won’t be able to maintain the velocity and ability to innovate in the new world. That is what we are seeing today in product and design in addition to publishing,” said Annez. Watch the entire conversation on our [youtube](https://www.youtube.com/watch?v=4IFy_61N8q4) to get all the insights. Or dive deeper into how Braze cut the time to turnaround content by 98% with [AI-powered content operations.](/events/braze)