Launched this week
Turn coding agents into teammates anyone can use from Slack, Linear, CLI, API or your browser. Ship features, query data, build dashboards, automate workflows. All within your company's context, skills, integrations, and security guardrails.
Used Runtime as soon as it launched and it was super useful to get non-technical folks setup and contribute code to a mature codebase while having the right guardrails in place.
Runtime
@oscar_oliva glad Runtime has been of help! Thanks for the early feedback.
Sandboxed coding agents sound intriguing. How do they handle permission issues when running scripts on a shared server? Seems like a neat way to let everyone experiment safely.
Runtime
@andrewbuilds Each agent runs in its own isolated sandbox snapshotted from a pre-configured template, so there's no bleed between users. Secrets go through an encrypted vault and proxy, agents never see raw credentials. Admins can also set egress allowlists, command-level gates, and spend caps, so experimentation stays safe without anyone stepping on each other.
In conclusion: Runtime avoids the shared server problem entirely. Each agent runs in its own isolated sandbox, so there's no permission collision between users.
Good job guys, I think it should be clearer how the whole flow works and if at any point you guys have access to our repos and if so how do you manage it, I understand it's sandboxed and you delete the filesystem afterwards but it's not entirely clear if you have access somehow so that made me hesitate to use it
Runtime
@jose_cyc thanks for the feedback! How do you think we can clarify it? Would SOC2 compliant give you that trust you need to try us?
The "one agent expert whose setup nobody can reuse" is the most accurate description of where most eng teams are right now. Curious how you handle agent quality drift, when Claude or Codex updates under the hood, a packaged agent that worked last month starts producing different output. Is that the customer's problem to detect or does Runtime surface regression signals?
Really like the sandboxed approach here. Giving coding agents isolated environments feels much safer and more scalable for teams than shared setups. Curious how you manage resource limits for long-running agent workflows.
the company context and skills layer is the part that differentiates this from just wrapping an API but it's also the part that takes the most setup to be useful. how long does it actually take to get Runtime to the point where it knows enough about your stack, your data schema, your workflows to give useful output rather than generic responses? that onboarding investment is usually what kills team AI tool adoption before it starts
Embedding actual code execution inside Slack is a much bigger deal than it sounds, most "AI in Slack" tools just generate text and call it a day. The sandboxed environment is what makes this viable for teams, since nobody wants an agent touching production. Curious about long-running tasks though; if a test suite takes 8 minutes to run, does Slack handle that without the bot timing out?