For most of the last decade, security teams could set the pace of adoption. Say no, or say yes slowly. That worked when the technology under review shipped on a quarterly cycle.
Generative AI ended that. According to Sahil Agarwal, co-founder and CEO of Enkrypt AI, which Anaconda acquired in August 2026, the pressure is now coming from above instead of from engineers. “Security teams are being forced to move fast by their board, by their executive team,” Agarwal says. “They’ve already decided that we will be using AI across the business, and you [security team] figure out how to secure us, because you’ll be the one answering for it.”
He has watched the alternative play out. “In 2023 and 2024, a lot of CISOs were saying I’ve blocked all ChatGPT, I’ve blocked all AI within my organization. And that didn’t go so well with the businesses, or with the CEO, or the board.”
Enkrypt’s pitch inverts the role. “Rather than you being a blocker to innovation, we enable you to be positive reinforcement for innovation in your business. If you use us, you can literally move fast.”
Supporting that promise is what shaped Enkrypt’s operating model: a twenty-person team, no product managers, and a research team that turns newly published attack techniques into product tests.
Seeing the Problem Before the Market Did
Enkrypt is an agent security and compliance platform with three offerings: red teaming to test agents for vulnerabilities before production, guardrails to protect them against attacks once they are live, and compliance to verify they meet federal, state, and local regulations alongside internal policy. It covers LLMs, chatbots, copilots, MCP applications, and agents themselves.
The company started in July 2022. ChatGPT launched that November.
“We knew this problem was coming, based on our own experience of how AI is moving forward, and what problems the whole industry would be facing,” Agarwal says.
Enkrypt runs an active research team that publishes at academic conferences and reads the field continuously, folding new methods into the product as they appear.
“If you’re not innovating on the product, then the company is already dead, because the field has already moved forward.”
Building Is Not the Blocker
Enkrypt’s development process starts from customer demand, or from its own research, and moves almost immediately to working code.
“It’s very much: I have this idea, let’s build a prototype, see how it functions,” Agarwal says. “In today’s day and age, building is not a blocker. You don’t need to spend time deliberating, you don’t need to spend time mocking up how it would look. You can literally build it in a matter of hours.”
The prototype then goes out to existing customers for a read on whether it matters. If it does, the team scales it up. If it doesn’t, they move on.
Asked about product managers and approval chains, Agarwal answers: “We don’t even have that. We are the product manager and we are the engineers.”
At Enkrypt, engineers are expected to chase their own ideas, even when most won’t work out. “You should be making mistakes with ideas. You should be trying out as many things as possible. Otherwise Prashant or I, as the co-founders, will not have all the ideas out there.”
Prioritization is equally direct. The team asks whether there are customers lined up who will pay for something, and how quickly they need it.
Scanning 25,000 MCP Servers
The clearest demonstration of what that operating model produces came in the two months before the acquisition. Enkrypt built an MCP scanner, then pointed it at every publicly available MCP server it could find across GitHub, npm, and other public sources.
The result: more than 268,000 tools scanned across 25,000 MCP servers, surfacing more than 143,000 vulnerabilities affecting 73% of those servers.
The findings went back to the community. “Security needs to be a community-driven thing, where you’re sharing vulnerabilities,” Agarwal says. “A rising tide lifts all boats.”
It also created a market. Customers now run the MCP scanner in production, and other companies license Enkrypt’s MCP vulnerability data to assess third-party risk for their own customers. “Ultimately, no one wants supply chain risk as they’re building these agents.”
What This Produces
Shipping cadence: Two to three features a week, and faster when a customer needs it. Agarwal describes shipping in under a day when a demo was on the calendar.
Recent releases: In back-to-back weeks: an MCP gateway, then a full platform update with a new UX and an agent skill scanner.
Category firsts: Compliance-based red teaming, when competitors were still doing static testing. The MCP scanner, launched as the first of its kind. And scanning for agent skills.
Dogfooding at Anaconda Scale
Enkrypt secures its own MCP server with its own tooling. “We understand that aspect, so we can make it a lot more secure.”
The more interesting work is just beginning. The team is integrating Enkrypt’s guardrails into Anaconda’s internal environment, which means moving from a 20-person company to a 400-person one.
“When we’re selling to an enterprise with 10,000 employees, we want to understand how people are using it at that scale. What are the guardrails? Going to the next level of dogfooding, at Anaconda scale, is something that’s actively being worked on.”
The Discipline Behind the Speed
Speed at this pace has a cost, and Agarwal is direct about where it lands. Things break. Early releases are not as stable as an enterprise customer would want.
His answer is a ratio. “Seventy percent of your time is focused on your main job. Thirty percent is where you’re experimenting. The thirty should not break the seventy, and vice versa.”
The other cost is organizational. “We’ve had situations where the engineering team is building so fast that marketing is struggling to catch up. We launched the MCP scanner today, then the agent scanner tomorrow, and people are asking what is even happening.”
“The metric is not how fast you’re shipping something,” Agarwal explains. “The metric is how many experiments you’re running, to make sure you’re making the right thing.”
Building AI your security team can approve? See how the Anaconda Platform governs every step from a builder’s first prompt to production. Talk to our team.
Anaconda Platform
The Only Unified Platform for Open Source