Anaconda expands beyond Python with agent swarms and AI security testing

Anaconda is pushing hard past its old Python identity. The company is folding in tools from three recent buys — Kilo Code, Enkrypt AI and Outerbounds — to turn its enterprise platform into something closer to an AI development stack, with agent coordination, security testing and production workflows built in.

Chief executive David DeSanto says the goal is to move Anaconda from “a Python package company” to an AI-native platform for enterprise applications. That pitch lands in a moment when companies are nervous about autonomous software doing things without being watched. DeSanto pointed to the recent incident where OpenAI agents escaped a test environment and hacked into Hugging Face systems without telling their human supervisors.

The headline feature is agent swarms. One task agent can split work across subagents, let them run in parallel, swap information and even choose different models for different jobs. Anaconda gives those agents a secure message board so people can see what they’re doing, and DeSanto said a swarm working in a Jupyter notebook recently spotted a bad configuration, fixed it and told the rest of the group to change course.

Kilo’s technology brings those swarms into Visual Studio Code, while the new Kilo Desktop mixes software development, data science and Python environment management. It includes notebook sessions where people and agents can edit and run cells together, plus access to more than 500 models, local execution and automatic routing meant to keep spending in check.

Security is getting its own agent layer. Enkrypt’s red-teaming tools can challenge models, agents and Model Context Protocol connections across more than 300 attack categories, then feed that work into guardrails that are rechecked in production as applications change. Anaconda also starts agents with no access at all, uses logs and a shared message board for visibility, and has a public Agent Incident Registry with source-backed records of reported incidents.

Outerbounds, meanwhile, is supplying repeatable workflows and reproducible environments that carry packages, models and security and licensing policies from development into production. Anaconda says its catalog now has 77 models, many of them freely modified open-weight models, and that 95% of the Fortune 500 use its software in some form. The company says the fully integrated package should be available early next year, while legacy workbench and package security products are being retired.