For the better part of this decade, I viewed Developer Experience (DX) as the North Star of data platform engineering. We obsessed over making platforms that were easy for people to navigate. We built intuitive developer portals and golden paths. We assumed the end user would always be a human sitting at a desk, looking at a screen.
My work over the last six months has forced me to rethink everything.
The developer we are building for is no longer just a human. As we move deeper into 2026, it is clear that AI agents will do the heavy lifting in coding and system management. If we continue to build for human eyes only, we will inevitably become the bottleneck in our own systems.
We need to move beyond DX and start engineering for Agent Experience (AX).
Why the shift to AX is non-negotiable
For an agent to be truly autonomous, it needs a system that is perfectly structured, predictable, and programmatically accessible. A portal that feels intuitive to a human may be a labyrinth for an agent. While a human can “guess” their way through a slightly ambiguous UI, a machine requires high-precision control and context.
A great platform in 2026 will be judged by how well an agent can understand it without a human in the middle to translate.
The technical pillars of agent experience
Building for agents requires a fundamental shift in our technical priorities. Here are the five areas where I believe we need to focus our engineering efforts.
1. Programmatic over visual
We must move away from visual interfaces and toward headless, API-first environments. An agent needs a CLI or a robust API, not a button on a web page. This realization was the driving force behind a major strategic pivot we made regarding our data catalog this past fall. We stopped prioritizing the search UI and started prioritizing the machine-readable metadata layer.
2. Structured schemas over prose
Documentation needs to be machine-readable by default. Instead of long tutorials or sprawling Wikis, we need strict APIs, technical specs, and metadata files. These files must explicitly define what a tool can and cannot do in a way that an LLM can parse instantly without having to “read between the lines.”
3. JSON/YAML as the primary language
Communication must be deterministic. Structured JSON with strict schema adherence ensures that an agent can pass instructions and data without the risk of misinterpretation. This is why we are increasingly favoring CLI flags like --json to ensure the output is always predictable and ready for the next agent in the chain.
4. Recovery paths in error messages
A standard error message is a dead end for a machine. True AX means providing structured error hints. If a request fails, the system should provide a reason that allows an agent to self-correct and retry without human intervention. We are moving from “Error 400” to “Error 400: Schema mismatch on field X, expected type Y.”
5. Autonomous discovery standards
We need universal standards that allow agents to discover and use tools and data across different systems. This is particularly challenging in complex environments where customers may have multiple repositories to manage, even if the primary solution is maintained in a monorepo. An agent must be able to navigate this multi-repo complexity to build pipelines and perform analysis autonomously.
Are We Ready for the Machine User?
The transition from DX to AX is not just a technical change; it is a cultural one. It requires us to stop being designers of user journeys and start being architects of agentic protocols.
I am curious if other platform engineers are starting to build with AX in mind. Are we ready to pivot our platforms to serve the machines that are now building our software? How are you actually building for AX today?
I’m committed to keeping this content free and accessible for everyone interested in data platform engineering. If you find this post valuable, the most helpful way you can support this effort is by giving it a like, leaving a comment, or sharing it with others via a restack or recommendation. It truly helps spread the word and encourage future content!
You can also connect with me or follow my profile on LinkedIn, where I share shorter-format thoughts and daily reflections on data platform and agentic engineering.
Full Disclosure: I am a Google Developer Expert, but I am a Google customer first. I maintain a critical eye when exploring and evaluating Google’s services.


Hello Robert
I'm an editor at ISMG and writing a series of articles on Data management for our online publication.
Would you be available to do an email interview?
I'm on LinkedIn too.
Regards
Brian Pereira