Modern analytics in the age of agents

Omni has positioned itself as the simplest modern analytics platform — consolidating governed BI, ad-hoc SQL, and spreadsheet workflows into a single tool. At HumanX 2026, WorkOS CEO Michael Grinich sat down with Omni CEO Colin Zima to talk about where analytics is headed now that AI agents are part of the picture.

Customers come to Omni for different reasons. Some want to stand up an analytics agent from scratch. Others are consolidating scattered tools. Many just want one platform that replaces three or four point solutions. But across all of them, the same question keeps coming up: what changes when agents enter the workflow?

Agents are just instructions and a cron

Zima offered a refreshingly pragmatic take on the agent hype:

"I literally think an agent is just a set of instructions and a cron. Maybe that's a gross oversimplification."

In Omni's world, AI started as an accelerator for existing human workflows — writing queries faster, building dashboards faster. Now it's evolving toward light automation: monitoring product areas, running analyses on schedule, following documented instructions.

But the key word is lightly. These agents are still heavily human-in-the-loop. When asked why, Zima pointed to two factors: risk aversion and the quality of results. For well-defined tasks, agents can run autonomously. For ambiguous analytical questions — "why did this region underperform?" — AI gives you a response, but you often want to correct and guide it.

Autonomy scales with task specificity.

The three-layer future

Zima laid out a vision for how software interfaces evolve from here. Everything will have three layers:

Dashboards become another surface for this pattern. But Zima doesn't think AI replaces UI. AI accelerates UI.

"There are so many workflows in business software that are 25 clicks where eight words is probably the simpler way to do it. But you still want clicks to refine."

Natural language is a good starting point, but precision still requires direct manipulation.

The vibe-coded piano problem

On the topic of generative UI and disposable software, Zima offered a vivid analogy. Imagine vibe-coding a piano where the keys change position every time you sit down. You'd never learn to play it.

Analytics has the same property. There's real value in looking at the same dashboard over time — consistency matters for building intuition and catching anomalies. If the interface regenerates itself every session, you lose the muscle memory that makes experienced analysts fast.

"I've jokingly tried to use AI for Photoshop, and when you're trying to keep this part but tweak that one thing, it doesn't do quite as good a job of fine-tuning. There's still a place to inject UI between AI."

Generative UI is powerful for exploration but dangerous for monitoring. The best tools will know the difference.

80% AI-generated code (and climbing)

Zima surveyed his engineering team and found that self-reported AI assistance ranges from 80% to 99.9% of their code. The remaining human contribution is judgment, correction, and curation.

"The curation becomes more important. But ideation and the first cut? You can do that 10x faster."

This tracks with what engineering teams across the industry are reporting in surveys, though self-reported figures like these are difficult to measure consistently. The bottleneck is shifting from production to evaluation. Writing code is cheap. Knowing which code to keep is the skill that compounds.

Build more, throw more away

For running the company, Zima's approach is straightforward: build more aggressively and be willing to throw things away.

He shared an example. Two engineers built a Microsoft Paint–style drawing feature on top of Omni's dashboard canvas in a day and a half during a hack day. Previously, that kind of experiment would have taken weeks of planning and dedicated engineering time.

"You can prototype things into prod a lot more aggressively than before. You just still have to be willing to curate."

That willingness to discard is the cultural shift AI demands. When the cost of building drops dramatically, the cost of keeping the wrong thing becomes the dominant risk. Teams that internalize this — prototype fast, evaluate ruthlessly, ship only what survives — will pull ahead.

This interview was recorded at HumanX 2026 in San Francisco.