Inside IntelliStream. How AI agents work alongside us
We are committed to building a data platform to extract significant value from operations. This means we have to adapt to new ways to work, and use new technologies for increased performance and quality. Lately we changed significantly in the way we work with code. AI Agents have quietly become our new colleagues.
We wanted to write this down because part of how we want to work with our community is to be honest about what we're doing and how. If you're going to trust us with your operational data, you should know what's behind the curtain.
Here's what AI agents do for us, in practice.
Agent Orchestration
A lot of our work isn't one task, it's a sequence. Understanding requirements and writing it down. Create tasks, look at the related code, draft a fix, run the tests, write a pull request and post it for review. By using agents this chain runs faster than any of us could pilot it manually. One agent picks the work, others execute it, and a final one summarises what happened. We supervise; we pilot, and review every action.
They help us build applications faster
From propping a new service to writing the first cut of a UI, agents shorten the distance between "we need a thing that does X" and "the thing exists, here are the rough edges to polish." We still own the design decisions and the production readiness call. We just spend less of our time on the boilerplate that surrounds them.
They automate the operations that used to eat our Fridays
Platform upgrades. Rollbacks when an upgrade goes sideways. Data migrations across versions. These are the unglamorous tasks every small team puts off because they're tricky, faulty, and hard to delegate. Agents can rehearse them in a sandbox, run them under supervision, and document what they did. For a team our size, that's a serious improvement.
They let us replace proprietary tools with internal ones
We have several small internal applications with important features that we need instead of paying SaaS vendors, giving us flexibility, data sovereignty and that fits our workflow. Building a focused internal tool used to cost weeks of a developer time. Now it costs an afternoon. We can continuously optimize and broaden our line of work.
They speed up research, education, and training
When a new technique appears, an agent can read the paper, summarise the practical implications, draft a small experiment, and tell us where the challenges are. New team members can use them the same way, as an always available colleague who's read every line of the codebase, ready to explain the corners the docs haven't caught up with yet. Now, the senior engineer's brain will not be the only place where certain knowledge lives.
They help us build data pipelines faster
Plumbing data from a new source into the model used to be a week of work. With an agent that understands our ontology and writes the connector code under review, the same job lands in a day or two. That speed flows through to us with faster onboarding for new sources and fewer surprises in the pipeline.
They run our testing infrastructure, partly on their own
Our test environment can spin up scenarios, run them, triage the failures, and surface the ones that look like real regressions. We still investigate the interesting failures ourselves. The tedious ones, silos, environmental noise, known resolved issues, get filtered before we see them. The result is a testing layer that catches more, complains less, and doesn't need a human babysitting it overnight. And if a bug is found, a new issue is created, another agent writes the pull request and the human senior developer reviews it.
The point isn't replacing engineers. It's automating the plumbing.
There is a long tail of hundreds of boilerplate tasks, integrations, migrations, and edge cases that each need an hour of attention. Agents are unreasonably good at that long tail. They give us back our time to spend on the parts of the platform that need actual judgement and which decisions are worth at which cost, and what's the next thing to build.
We're sharing this because we want you to be part of our journey as we build IntelliStream. The platform under your data should never be a black box and neither should the team behind it.
If you're working on similar problems, or if you have questions about how any of this fits with our platform, write to us. We would love to hear from you.