Agentis Blog
Product thinking, engineering notes and real stories about turning AI coding agents into reviewed, shippable work.
Chat logs are where delivered work goes to die. Here is why we built Agentis around tasks, approvals and artifacts instead of conversations.
Ondřej Novák
Founder
AgSpec turns an ambiguous task into four linked artifacts: scope, observable requirements, technical design and a verifiable implementation checklist. The result is a contract that one agent can write, another can implement and every later run can resume without guessing.
Ondřej Novák
Founder
Type @agentis in a PR comment — or on a single line of the diff — and an agent run starts with the whole pull request as context. A look under the hood of the Agentis GitHub bridge: webhooks, signature checks, inline code locations and the guards that keep a noisy repo from launching a fleet.
Ondřej Novák
Founder
Mention the bot and a task is born; the agent streams its thinking back into the thread and asks you questions with real buttons. A look under the hood of the Agentis Slack bridge — events, guards, streaming and an interactive MCP.
Ondřej Novák
Founder
Environment setup, the agent itself, commits, pull requests and cleanup — Agentis workflows move the entire run pipeline into a declarative file that lives in your repo.
Ondřej Novák
Founder
A manual-style reference for .agentis/workflows: which file runs when, what [%TOKEN%]s and env variables each step gets, the exact if-condition grammar, output types and extends merge rules.
Jakub Černý
Platform
A mention in Slack or a comment on a pull request is one way to start an agent. Your own code is another. The Agentis SDK is a Python client, a CLI and an MCP server over one JSON-RPC endpoint — create tasks, read runs, and block on a human answer, straight from a script, a CI job or a cron.
Ondřej Novák
Founder
A walkthrough of the full Agentis loop: a customer report in Slack becomes a tracked task, an agent run, a reviewed diff and a deployed fix.
Petra Svobodová
Product Engineering
Per-task, per-team and per-model cost visibility changed how our pilot teams budget for AI. Here is the model we settled on and why.
Jakub Černý
Platform
Autonomous agents that never ask are a liability. How we designed the question and approval flow so humans stay in charge without slowing runs down.
Marek Dvořák
Backend
Knowledge extraction turns completed runs into release notes, decisions and reusable context. The compounding effect is the real ROI of agent platforms.
Ondřej Novák
Founder
Claude Code today, OpenCode tomorrow, your own adapter next quarter. Why agent-runtime portability is an architecture decision, not a checkbox.
Jakub Černý
Platform
Stay in the loop
One email a month with our best writing on running AI agents in production. No spam, unsubscribe anytime.
