How to give agents autonomy safely, which language makes it easier to review what AI writes and a step forward for using AI on sensitive data.
Docker Sandboxes isolate coding agents in disposable microVMs
Docker launched Docker Sandboxes, microVM-isolated environments for running coding agents unattended. Each sandbox has its own Docker Engine, with no privileged access to the host, and the organization decides beforehand which files, endpoints and secrets the agent can use. A sandbox can move between laptop and cloud with one command, for heavy builds or so the agent keeps working with the laptop closed, and can be deleted and recreated in seconds.
Giving agents more autonomy is safer when the boundary is infrastructure, not only prompt instructions. Teams can standardize one agent environment for every developer, with access rules set by the company.
How Dexpert helps: Cloud & ServerlessSources
Google makes the case for Go as the ideal language for AI-assisted engineering
In a Google post, the Go team argues that when agents generate hundreds of lines in seconds, what matters is no longer how easy code is to write but how easy it is to review, verify and maintain. The post recalls that Go was created for team software engineering: simplicity with one way to format and test, strong compatibility guarantees, a global dependency system and built-in security tools. For Google, these same qualities make it easier to work with AI as part of the team.
When choosing technology for new systems, weigh how easy the code will be to review and maintain, since much of it may be written by agents. Simple languages with standard tooling lower the cost of that review.
How Dexpert helps: Backend & APIsSources
- Google for Developers: Why Go is an Ideal Language for AI-Assisted Software Engineering
Google open-sources HEIR, a compiler to run AI on encrypted data
Google presented HEIR, an open source compiler that makes it easier to use homomorphic encryption, a technique that allows computation directly on encrypted data. With it, a cloud service can, for example, produce recommendations without seeing the user's data, and the answer comes back encrypted. Google notes the technique still has extra cost, but that this cost is falling fast, which matters to sectors such as healthcare and finance with strict data-sharing rules.
Using AI on sensitive data, such as medical records and financial information, may stop depending only on trusting the provider. Regulated companies can follow the technique for projects that today run into privacy limits.
How Dexpert helps: Artificial IntelligenceWant to put one of these to work in your company?
Dexpert builds custom systems, apps, integrations and AI solutions. Tell us what you need and get an estimate, or take the free digital maturity assessment.