Back to Features

Ask Chat for a competitor breakdown and you get a competent answer in one shot. Ask an agent and it goes wider: it searches across sources, reads the documents you connected, cross-checks what it finds, drafts the breakdown, and flags the parts it wasn’t sure about. Same starting prompt, very different ending. The difference isn’t intelligence — it’s that an agent is allowed to take steps.

That’s the line worth drawing. Chat is conversation. You think out loud, get answers, refine. An agent is delegation. You hand off a goal and a bit of context, then walk away while it works through the parts. One is a tool you hold. The other is a worker you brief.

The range of agents

Skynet gives you a roster, each tuned for a different shape of work:

• Research — gathers, reads, and synthesizes across your sources and the web, with the trail of where things came from.
• Writer — drafts, edits, and rewrites in your voice, using context from your own documents instead of guessing.
• Code — reads a repo, makes changes across files, and opens a pull request you can review.
• Designer — turns a brief into visual work, like a storyboarded video ad from a few lines of direction.
• Analysis — pulls numbers together, finds the pattern, and explains what it actually means.
• Assistant — the generalist for the odd jobs that don’t fit a neat box.
• Memory — keeps your context current so every other agent already knows what matters.

You don’t pick from this list like a menu. You describe what you want done, and the right capability gets put to work.

When to reach for an agent

Stay in Chat when you want a fast answer, a second opinion, or a back-and-forth where you’re steering every turn. The loop is the point.

Reach for an agent when the job has steps you’d otherwise do yourself: anything that means visiting several sources, touching several files, or producing a finished artifact rather than a reply. A good tell — if you’d normally say “give me an hour,” that’s an agent. If you’d say “quick question,” that’s Chat.

Running one

Because Skynet’s integrations let agents take action, not just read, that final step is where delegation pays off. The agent doesn’t hand you a draft to go paste somewhere — with your approval, it posts the message, opens the pull request, updates the page.

Make it your own

The agents you run most are usually the ones you’d run the same way every week. That’s the cue to build a custom one. Describe the job once — the goal, the sources, the tools it should touch — and you’ve got a reusable agent you can fire on demand or put on a schedule. No code. The status report that ate your Monday mornings becomes a thing that’s already done when you wake up.