From BI reports to agents: what changed when everything started from AI
AI changed how fast things get built. It did not change what is worth building.
That line is the home-page thesis. This essay is the longer thread behind the timeline: how the work moved from static reports to systems that act — and why the business conversation never left the room.
The path in five steps
The home page timeline is chronological practice. Compressed, it is five stages:
- Data — trustworthy platforms before models matter. Pipelines, lineage, access. Case: enterprise data platform.
- ML and GenAI in production — forecasting, NLP, questions to SQL on real data. Case: AI analytics platform.
- AI automation — models wired into everyday workflows: collect, transform, decide, publish. Case: AI automation in production.
- Agents — roles, tools, memory, approvals, cost routing. Case: Autonomous content org.
- Business — what is worth building, for whom, at what cost. Cases: B2B evidence across many markets and TUVA.
Breadth is the point. The same person who founded a product and shaped platform investment also built lakehouses and agent stacks. That combination is the positioning — not a pile of tools.
What BI was good at — and where it stopped
Early work was ServiceNow history, SQL, and Power BI: visibility into operations, SLA trends, support patterns. That layer still matters. Reporting answers “what happened?”
It does not ship the next action. Dashboards die in folders for the same reason unused AI output does: no owner for the decision, no workflow for the follow-through. I wrote about that shift in from unused dashboards to unused AI output.
Platforms made models useful
GenAI without governed data is a demo. The useful step was a lakehouse the business could trust — jobs in production, catalogued tables, NLP on support history — then natural-language access that still validated SQL before anyone trusted the answer.
Speed without lineage is just faster wrong answers.
Automation, then agents
Once models were reliable enough, value moved into the workflow: translation pipelines, Slack-to-site publishing, recurring digests, approval gates. Orchestration tools changed (n8n earlier; Inngest for many recurring flows now), but the shape stayed the same: deterministic checks around generative steps, humans on the irreversible ones.
Agents are that idea with roles and memory. The hard part is not autonomy. It is access, evals, spend, and who may publish. That is agentic engineering in practice — and guardrails when you mean it.
The constant
Every step asked the same question TUVA and B2B platform work taught early: what is worth building? AI made the build cheaper and faster. It did not remove the need for evidence, cost, and an owner of the outcome.
If you only need the map, start on the home page. If you need the systems, start in projects.