How AirOps chases friction to build AI products with Claude

This should update to: Claude solved execution but shipping agents into a specialized professional workflow like content marketing surfaces two harder problems: making agents fit the workflow marketers already run, and engineering a harness that clears the enterprise quality bar. Claude Managed Agents and the Claude Agent SDK let us iterate fast enough to hit that bar, and this talk shares the opinionated lessons from building AirOps Next (Quill, Playbooks, Inbox) on top of them.

May 22, 202626mWatch on YouTube ↗

WHAT IT’S REALLY ABOUT

AirOps reduces marketer friction with Claude-powered agents and playbooks

AirOps shifted from a complex, brittle node-based workflow builder to an agent-and-playbook approach to better fit how marketers actually create and review content.
The company launched AirOps Next with Quill (an agent “captain”) and Playbooks (document-based, collaborative, versioned skills) to move users from insights to actions faster.
To prevent “endless use cases” sprawl, AirOps scoped agent design around real marketer workflows and inserted explicit human-in-the-loop review gates for governance.
To improve output consistency, AirOps invested in “harness engineering,” especially specialized tools for efficient context gathering and sub-agents for focused tasks like compliance, writing, and brand context retrieval.
Next challenges include building self-improvement/feedback loops (memory, trace summarization, forgetting) and benchmarking creative content agents where “correctness” is taste-dependent.

IDEAS WORTH REMEMBERING

5 ideas

Agents become harder—not easier—when the user is a marketer, not a developer.

AirOps found that concepts like JSON, variables, and brittle step dependencies create a “complexity ceiling” for marketers, pushing them toward a more familiar document-based authoring model.

Constrain the problem space to avoid agent product sprawl.

Because agents can do “everything,” AirOps forced intentionality by anchoring design to a concrete marketer workflow (discover → research → brief → draft → optimize) with defined review points.

Document-based playbooks preserve approachability while still enabling transparency and control.

Playbooks behave like collaborative SOPs/skills with explicit inputs/outputs and tool references, helping users understand what the system is doing without a node graph.

Governance is a first-class feature: human review gates are built into execution.

AirOps assigns approvers at section checkpoints, routes approvals via an inbox, and supports review at scale in a grid—so the agent can’t proceed until the right person unblocks it.

Consistency comes from harness engineering, not just choosing a better model.

They frame Claude as the “engine,” but emphasize surrounding components—especially tools and context orchestration—to reliably reach enterprise-quality outputs.

WORDS WORTH SAVING

5 quotes

I guess the main really big takeaway I want you guys to come away with is building agents, um, and just making agents accessible is honestly a really hard problem.

— Dylan

With a workflow builder style, and especially with our, our core customer audience being marketers, um, you would hit this complexity ceiling where you're trying to teach a content marketer what liquid text is, what JSON is, and all these, you know, different concepts.

— Dylan

It's really easy to start sprawling into this spiral of, um, yeah, just like there's so many different use cases.

— Dylan

I think a lot of it is w- like, with agents, also with coding, is you kind of let it go.

— Dylan

Every single time a problem is solved, just that friction point always keeps moving.

— Dylan

AI search (AEO) as “SEO for ChatGPT/Claude/Gemini”Workflow-builder complexity vs. agentic experiencesAirOps Next launch: Quill and PlaybooksDocument-based playbook/skill authoringHuman-in-the-loop governance (assignments, inbox, review gates)Harness engineering: tools, context orchestration, sub-agentsToken efficiency, speed, and enterprise time-to-value metricsFuture work: feedback loops, memory management, benchmarking creative quality

High quality AI-generated summary created from speaker-labeled transcript.

Get more out of YouTube videos.

High quality summaries for YouTube videos. Accurate transcripts to search & find moments. Powered by ChatGPT & Claude AI.