A small team,
with an unusual teammate
Cloud Spectra is built by a networking veteran and by Claude -- an AI engineering partner from Anthropic that works in this codebase every single day, under exactly the same rules as everyone else.
Ananth Suryanarayana
Twenty-five years in data networking, 20+ USPTO patents in software-defined networking, and two IETF RFCs. Sets the architecture, owns the roadmap, and signs off on every release that reaches a customer account.
Claude
Claude is not a tool we occasionally reach for -- it is a working member of this team. It reads the whole repository, writes and reviews Go services, Angular components, CloudFormation, and Terraform, hunts regressions across 18 binaries, audits IAM policies condition key by condition key, and keeps the docs and this website honest with the code.
It also does the unglamorous parts nobody volunteers for: sweeping a one-line bug fix through every service that shares the pattern, writing the failing test before the fix, and reading the logs before guessing. A great deal of what makes this product careful is work Claude did at 2am without being asked twice.
A team this small should not be able to ship an appliance with this much test surface behind it. That it can is the whole point.
The same engineering rules apply to every contributor on this team, human or not.
Tests come first
Every change lands with unit tests. Every bug fix ships with a test that fails on the old code and passes on the new one -- plus a written explanation of why the old suite missed it.
Fix it everywhere
A defect found in one service is treated as a defect in every service that shares the pattern. The sweep happens in the same change, not in a follow-up ticket that never comes.
Least privilege, verified
Every IAM condition key is checked against the AWS reference for that exact action and resource type before it ships -- because a key that is not valid there silently evaluates to null and the grant quietly disappears.
Read the logs first
When something breaks in a live deployment, the first move is the log line and the journal -- not a packet capture. Most root causes announce themselves in plain text at the moment they happen.
Docs move with the code
The user guide, the deploy page, and the generated schema are part of the change, not an afterthought. Drift between them fails the build.
Nothing ships unreviewed
Claude proposes and writes; a human reads every diff, runs the full suite, and decides what becomes a release. The tests are the arbiter, and a person is accountable.
Where the line is
We are deliberate about this, because you are trusting our software with your network. Claude works in our repositories and our test accounts -- never inside a customer's AWS account, and never with customer data. The gateway you deploy runs entirely inside your own AWS account.
Every change follows the same path regardless of who wrote it: unit tests, integration tests against live AWS in both the CloudFormation and Terraform deploy lanes, a human review, and a signed release. If a change cannot survive that gate, it does not ship.
And yes -- we ship Claude to you, too
The same model that helps us build the gateway is a first-class route inside it. The AI Gateway's LLM proxy fronts Anthropic Claude alongside other providers, with response caching, per-team keys, and usage accounting -- so your inference traffic gets the same flat-fee, zero-per-GB treatment as the rest of your data plane.
Want to build alongside us?
Come argue about packet paths, IAM conditions, or AWS bills in our community Slack.