How to Give AI Agents Enough Context to Be Useful

Skyvern is an open source company helping healthcare businesses automate manual browser tasks, and co-founder Suchintan Singh scaled it past $2 million in run rate while personally handling PM, sales, marketing, and customer support — all powered by AI agents. In this recent batch talk, Suchintan breaks down why agents produce slop when they lack business context, walks through two specific skills he built, and explains why he banned DMs and records every call to make his entire company legible to AI.

Suchintan Singhguest

May 19, 20265mWatch on YouTube ↗

WHAT IT’S REALLY ABOUT

Give AI agents full business context to avoid slop output

Agents fail when they follow instructions without the business context needed to make good decisions, much like a new employee without onboarding.
High-quality context comes from giving agents access to the company’s living knowledge base—emails, Slack, Notion/docs, call recordings, and sometimes production data.
Skyvern’s PRD-writing agent searches across recorded customer and internal conversations, drafts a spec grounded in evidence, then uses adversarial review and a prioritization framework (RICE) to remove weak requirements.
A daily content-marketing agent turns the last ~20 customer conversations into targeted post ideas and drafts, enabling consistent publishing and occasionally generating unexpectedly high-value, specific leads.
Remote-first workflows naturally create more “capturable” context, while in-person cultures must intentionally record and centralize knowledge to unlock agent performance.

IDEAS WORTH REMEMBERING

5 ideas

Agents are only as good as their onboarding data.

The speaker frames poor AI output as the natural result of missing business context; supplying the same background a human would need is the fastest path to usefulness.

Treat internal tools as a single knowledge base—and record what matters.

Slack threads, Notion pages, and call recordings collectively become “company memory”; unrecorded hallway conversations and DMs create blind spots for both humans and agents.

Ground PRDs in searchable evidence, not just prompts.

Skyvern’s PRD agent searches across call recordings, Slack, docs, and customer communications, then links relevant sources so engineers can verify the rationale (e.g., CAPTCHA-solver complaints).

Add structured critique and prioritization to reduce hallucinated requirements.

Sub-agents perform adversarial review and leave comments, and a framework like RICE is applied at the end to cut “junk requirements” the model tends to invent or overweight.

Automate content ideation from real customer conversations to stay specific.

A daily email converts recent calls into buckets like recurring pain points and contrarian takes, producing platform-specific drafts that the human reviews before publishing.

WORDS WORTH SAVING

5 quotes

One of the big reasons that agents produce slop, which every engineer who doesn't like AI likes to tell you, is because they love to do what they're told, but they often don't have the context required to do it well.

— Suchintan Singh

Agents are the same thing. And so giving good instructions, giving good context, and letting them critique their own work is how you make them much more effective at their job.

— Suchintan Singh

And this is some place where remote companies have an unfair advantage. You know, every in-person company, the context is spoken.

— Suchintan Singh

So what we did at Skyvern was basically we stopped letting people send DMs. Nobody's allowed to send DMs in, in our company.

— Suchintan Singh

We record every call, internal or external. Even my one-on-ones with my co-founder are recorded, which, you know, whether they should be or not is an open question.

— Suchintan Singh

“Slop” as a context deficit problemCompany knowledge base as agent fuelTooling sources: Slack, Notion, email, call recordings, databasePRD generation workflow with evidence and reviewAdversarial sub-agents and self-critique loopsPrioritization frameworks (RICE) to prune requirementsRemote vs in-person information capture; anti-DM policies

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.