The idea
behind the loop
Coupa's DSLM work and Navi as a whole come down to the same thing: agent loops that retrieve, reason, and act across procurement, finance, and supply chain, and do it dependably. With Snowflake Cortex Agents, each leg of that loop runs next to the data, inside the same governance boundary, from the first experiment to production.
Read the Cortex Agents docsCortex Search + Cortex Analyst
A Navi agent grounds itself in spend data. Cortex Search pulls passages from contracts and invoices, and Cortex Analyst turns questions into SQL over your tables.
Agent orchestration
An LLM plans the task, breaks it into steps, and decides which tool to call next, following the instructions you write into the agent spec.
Tools + response
The agent runs the tools, checks the results, and returns an answer, all under the same role-based access controls as the rest of your account.
A demo you can
actually play
An agent loop retrieves, reasons, and acts, then hands off to the next step. That is exactly a triathlon: three legs, and the whole thing falls apart at the handoffs. So instead of another slide deck, we turned your agent loop into three games you can play. The fastest way to understand a loop is to try not to break one.
Trusted data
Navi agents and Coupa's domain-specific models are only as good as what they're grounded in. Spend, supplier, and transaction data stay in one place, so every loop starts from the same source of truth.
Governed execution
When an agent acts on a PO, an invoice, or a supplier record, it runs under the same roles, policies, and audit trail as the rest of the platform. No side doors.
Path to production
The DSLM experiment and the production Navi workflow run on the same foundation, so a promising prototype becomes something you can actually ship to customers.
Ramana,
We've really enjoyed the conversations with you and the Coupa team over the past few months, especially as we've been digging into what the next phase of AI could look like at Coupa.
When I heard you describe agents as “LLM calls running in a loop,” I couldn't resist taking the running part a little too literally. So we put together something just for you - the ENGO glasses, plus a small Agent Loop Triathlon challenge.
There's obviously a more serious idea behind the joke. A lot of what we've been discussing around Coupa's DSLM development, and Navi as a whole, comes down to making those loops dependable: trusted data, governed execution, and a path from an interesting AI experiment to something you can actually put into production.
Hopefully the glasses are useful on your next run - and the game reminds you of our dedication to our partnership. Looking forward to continuing what we've started together.