I came to AI reluctantly.
For the first couple of years of this AI wave, I watched from the sidelines while much of the software industry declared that everything had changed. It was only after the models became stable and reliable - reliable in ways that I could test for myself - that I even considered building with them.
Some of that skepticism never left, and I think that's healthy. It shaped nearly every decision I've made while building Sonar AI, and with all of the noise in the NetSuite AI market right now, I thought it would be useful to write down what I believe and why. Consider it a plain statement of how I think about this work, collected in one place.
Where the Work Happens
I believe that the most important architectural decision in a NetSuite AI tool is where the work happens. Sonar runs inside NetSuite itself. It's a Suitelet, living in your account, running under your login and your role. If your role can't see a subsidiary, then Sonar can't see it either. The permission boundary that your administrators already maintain is the boundary that Sonar operates within.
That one decision carries a lot of weight. There's no replica of your data and no warehouse - no standing copy of your general ledger sitting in someone else's cloud. When a session ends, your data is exactly where it started.
It also means that Sonar works with the tools your team already trusts. When someone asks for an income statement, Sonar runs the real NetSuite report - the same one your controller signs - with full use of the report's filters and period presets. And when a question calls for raw data, the aggregation runs as SuiteQL directly inside the database, next to the data itself.
I wrote about this last month in How Sonar AI Works With Large Volumes of NetSuite Data. The short version is that the database does the math, and the model reads small, verifiable summaries. A single query pass can read up to 500,000 rows, and exports can stream straight to a file without ever passing through the model at all. In a recent gross-margin analysis, the model never saw a single transaction line - it saw 130 summary rows - and the analysis still covered every line that mattered.
The Security Certification Question
Security certifications come up a lot in enterprise AI conversations, so let me explain why Sonar doesn't have a SOC 2 report - and why I believe that it doesn't need one.
SOC 2 certifies how a vendor's systems handle your data. It exists because most tools take your data somewhere - into their cloud, into their storage, into a pipeline they operate. Sonar has no such systems. There are no Sonar servers and no Sonar database. There is no place where your NetSuite data comes to rest outside of NetSuite. The code runs in your account, on Oracle's infrastructure, under Oracle's certifications, governed by the roles your administrators already manage. There's nothing to audit, because your data never touches anything of mine.
The one external relationship in the loop is the AI model provider, and you hold that relationship directly, under your own API key and your own agreement.
Inheriting NetSuite's governance didn't stop me from adding more on top of it. Every change Sonar makes is written to an audit trail inside NetSuite - one that Sonar itself can't delete. Sessions can run in read-only or no-delete modes, and those restrictions are enforced at the tool layer, so they hold regardless of what the model does. Record changes are previewed as before-and-after diffs before anything is saved. Deleting a production record requires an explicit confirmation, and that confirmation is recorded verbatim in the audit trail.
There's even a privacy mode that pseudonymizes customer, vendor, and employee names before anything reaches the model.
What It Can Do
Sonar's range surprises people.
Financial analysis, administrative work, security and permissions reviews, SuiteScript development, data quality sweeps, month-end support... you name it. Sonar can also reach outside the account when the work calls for it, connecting to external resources through Web APIs - fetching documentation, checking reference data, calling third-party services.
And I'll be honest about the limits, because I think that honesty is part of earning trust. Truly cross-system analytics - NetSuite plus Salesforce plus a data lake - is a different problem, and an agent that lives inside one system isn't the right tool for that job.
Showing the Work
My continued skepticism about AI shows in Sonar.
Every SuiteQL query that Sonar generates is visible in the conversation, ready to be read and re-run. Results have to reconcile - when an analysis claims to explain a change, the components have to sum to that change, to the cent. And because Sonar runs inside NetSuite, it can run its own SQL and the standard reports in the same session and compare them.
Sonar also supports - and even promotes - getting a second opinion. It can package its own analysis, with the raw evidence and the formulas, and send it to a different AI model for adversarial review. In a recent engagement, that review caught two reasoning errors in Sonar's own work, and the corrections were published in the final report, verdict by verdict. I want an assistant that expects to be checked, because that's the posture I bring to my own work.
Built to Pivot
Building this close to the frontier can be painful. Models change fast, and the best practices you committed to in the spring can look dated by the fall. Changing direction is hard when a product has been architected around one model or one set of assumptions.
Sonar was designed to pivot quickly. New models are easy to support as they arrive. When best practices shift, they get folded in. And Sonar learns as you use it - it keeps field notes about your account, recording verified schema quirks and lessons from past sessions, so the second time it does something, it's faster and less wrong than the first time. Continuous self-improvement? Yeah, we're on it.
On Price
Sonar AI is free. All you pay for is the tokens, at your model provider's rates, under your own key. It's free, and it always will be, so there's no trial to start and no clock ticking while you evaluate it. Install it, connect a key, and use it for as long as you like.
For teams that want a running start, the Sonar AI Prompt Library has 317 professionally engineered NetSuite prompts across 15 categories - everything from margin bridges to permissions audits to BOM cost analysis - and 26 of them are free.
It Should Be a Joy
Sonar is fast, fun, and flexible. It supports exploration - the odd question, the hunch you want to chase - while providing a safe environment and guardrails to work within. You can wander, because the destructive operations are gated behind previews and confirmations, and everything is logged.
The fun comes from the work itself. Watching an analysis assemble in a couple of minutes - one that used to take all afternoon - doesn't get old. At least it hasn't for me yet.
Give It A Try
Whatever NetSuite AI tool you're evaluating - including Sonar - I think the test is the same, and you can do it in an afternoon.
Pick a period. Pick a question that matters to your business. Ask to see the queries, and ask how the results reconcile back to the books. Maybe ask for a second opinion. Then decide for yourself.
I built Sonar to pass that test every single time. That's the whole idea.
If you're interested in Sonar, I encourage you to reach out to the team at the Anchor Group, NetSuite and Ecommerce Specialists and request access to Sonar.