In March I shipped Awesome DevOps AI, a curated list of AI tools, agents, and resources built for DevOps, SRE, and Platform Engineering. The first version had 100 entries. Sixty days later it has 219. The repo has been featured in two newsletters, picked up by AI Engineer Pack, and is the most-starred thing I have ever shipped on GitHub.
This post is the changelog. What got added, what got removed, what the patterns are when you look at 100+ new AI tools in a single 60-day window.
What is in the list
Twelve categories. Each entry has a one-line description, a primary use case, and a link.
- AI Coding Agents for DevOps work (Claude Code, Codex, Cursor, Continue, Aider, and 14 more)
- Infrastructure as Code Copilots (Pulumi AI, Terraform AI, OpenTofu helpers)
- CI/CD Intelligence (failure prediction, flaky test detection, build optimization)
- Observability Copilots (Honeycomb Copilot, Datadog Bits AI, New Relic, Grafana LLM)
- Incident Response Agents (PagerDuty AI, Opsgenie LLM workflows)
- Security Automation (Snyk, Wiz, Lacework with AI features called out)
- Cloud Cost Intelligence (FinOps platforms with LLM features)
- Kubernetes AI (k8sgpt, KubeBuddy, Komodor with AI features)
- Chaos Engineering Agents
- AI-Native Monitoring Stacks
- MCP Servers for DevOps (the newest category; added in April)
- AI Engineer Career Resources
The 60-day trend lines
Three patterns stand out when you watch 100+ new entries land in 60 days.
Pattern 1: MCP servers are eating ad-hoc CLI tooling
The MCP Servers for DevOps category did not exist in March. It now has 23 entries. Filesystem, Postgres, Kubernetes, Terraform, AWS, GitHub, GitLab, Slack, PagerDuty, Datadog. Anything that used to be a CLI tool with a separate "make it AI-friendly" layer is becoming an MCP server.
The implication: if your team is building AI workflows for ops, you are increasingly composing MCP servers, not writing custom prompts. The composition layer is moving down the stack.
I wrote about how to build and audit MCP server configs in the MCP guide, and the builder tool is at /tools/mcp-config-builder.
Pattern 2: Coding agents are converging, then differentiating
The five major coding agents (Claude Code, Codex, Cursor, Continue, Aider) shipped surprisingly similar features in Q2: plan mode, hooks, MCP support, parallel agents, model routing. The convergence was rapid.
Then they started differentiating again. Claude Code leaned into long-context (1M tokens on Opus 4.7). Codex leaned into speed and price. Cursor leaned into IDE polish. Continue went open-source-only. Aider stayed minimal.
For engineering leaders, the takeaway is: pick by your team's primary workload, not by the brand. The benchmark in GPT-5 vs Claude Opus 4.7 shows why.
Pattern 3: AIOps platforms are quietly winning
Of the 100 tools I added in 60 days, more than 30 were features in existing platforms (Datadog, New Relic, Grafana, PagerDuty, Honeycomb) rather than standalone products. The market signal is real: incumbents have distribution. Startups have to be 10x better to peel customers off a platform that just added an AI feature.
For SREs evaluating tools, the rule is now: check whether your existing platform shipped the feature before you add a new vendor.
What got removed
Eight tools shut down or went stale in 60 days. Three were YC-backed startups. Two were open-source projects whose maintainers stopped responding. Three were "GPT-wrappers" that lost the wrapper margin once the underlying model dropped in price.
The removals are documented in the repo with a "discontinued" note. I do not delete them, because their existence is a useful data point.
How I maintain the list
Three rules I follow strictly.
- No pay-to-play. Nobody has ever paid for a placement. If they ask, I document the conversation in the issue tracker and decline.
- Personal vetting. I or someone I trust has used every tool in the list for at least 15 minutes on real work. Marketing pages do not count.
- Weekly review. Every Friday I scan submissions, validate links, and remove anything that 404s or has not shipped in 90 days.
That review is one of the highest-leverage hours of my week. It compounds because the list compounds: the more curated it is, the more people send useful submissions instead of spam.
Why I started it
The honest reason: I was getting asked the same question ten times a week. "What AI tools are people actually using for DevOps?" The answers were scattered across newsletters, X threads, and Slack DMs. A list felt obvious.
Two months in, the list answers the question for me. People send it to their teams instead of asking me directly. The list does the work.
Submitting
PRs welcome. Read the contributing guide first. The bar is real: I reject roughly 60 percent of submissions for being too early, too narrow, or marketing-heavy.
Receipts
- Tools listed: 219
- Categories: 12
- Stars and forks: see the repo
- Median additions per week: 17
- Median removals per week: 1
- Weekly maintenance time: 60 to 90 minutes
- Submissions rejected: roughly 60 percent
If you want a second opinion on AI tooling decisions for your platform team, that is the engagement I run.