What is an Intelligence Agent?
If you've scheduled a SQL Server Agent job, you already understand the foundation. Here's what's different now.
See the fleet →You already know this part.
SQL Server Agent Job
Schedule: Every night at 2:00 AM Step 1: Run stored procedure (validate data) Step 2: Execute SSIS package (transform) Step 3: Run T-SQL (load warehouse) Step 4: Send email notification On failure: Alert operator
LucidAgent Intelligence Agent
Schedule: Every night at 2:00 AM Step 1: Read bronze data (validate completeness) Step 2: Transform through Silver layer (clean, enrich) Step 3: Analyze with AI (find patterns, anomalies) Step 4: Generate findings in natural language Step 5: Post to Slack, email, update dashboard On finding: Trigger downstream agents
The structure is identical. The difference is Step 3 — instead of executing deterministic logic, the agent reasons about what it found.
What an agent is
An agent is software that pursues an objective: it watches defined inputs on a schedule, decides what matters, and produces an output you can verify. On this platform, every agent is published as a contract — what it watches, when it runs, what counts as a meaningful change, and what it delivers.
Most automation is a workflow: a fixed series of steps that code controls end to end. An agent is different — a model makes decisions inside the loop: which changes matter, how to rank them, what to say about them. LucidAgent runs both. Workflows move data through the pipeline; agents decide what is worth your attention.
Anything on this site described as an example is labeled as one. Everything else traces to a logged run.
Four levels of autonomy
Reactive
Run → Read → Process → Report
What your SQL Agent does today
Adaptive
Run → Read → Reason → Decide → Act
Adjusts behavior based on what it learns
Proactive
Watch → Detect → Initiate → Execute
Acts on conditions, not just schedules
Collaborative
Sense → Share → Coordinate → Compound
Multiple agents working together
Agents are more powerful in formation
A single agent replaces a scheduled job. A fleet coordinates many of them, and a shared memory layer lets later runs use what earlier runs recorded.
How agents connect to the world
Slack
Findings go where your team works
GitHub
Opens issues when a schema changes
Web Search
Research in real time
News APIs
Read the news so you don't have to
SMS Alerts
Pages the on-call operator