Onboarding Now Shows What the Agent Can Do, Not Just What It Sees
Reading your situation back to you isn’t the same as proving you’re worth trusting with it. Onboarding got reworked so the strategy step flows directly into a live signal-scan — instead of stopping at a diagnosis, new agents now take a concrete first action and show you the result. First-agent trial handling and a Product Hunt discovery path got fixed up alongside it, and Daily Spark (your recurring update digest) is more reliable about actually landing.
Drafting, Approving, and Publishing Are Now One Flow
This cycle’s biggest chunk of work: content drafts, engagement replies, and channel publishing used to live in three loosely-connected places, each with its own approval quirks. They’re now a single unified flow — one confirmation step whether you’re approving a post, a reply, or a manual publish, with markdown rendering, locked-preview merging, and artifact attachments cleaned up along the way. If you’ve felt like the drafting experience had rough edges depending on which screen you were on, this is the fix.
Switching Between Agents and Accounts No Longer Loses Your Place
If you run more than one agent or connect multiple X/Reddit accounts, identity handling got a real overhaul: agent switching has a proper UX with cross-agent notifications, X and Reddit now share one unified identity lifecycle, and OAuth fallback + approval context follow you correctly when you switch. A shared public-account exchange means teammates can reuse the same connected identity without re-authenticating from scratch.
The Agent Can Now Draw and Edit Its Own Images
Two new built-in tools: agents can generate images from a text prompt and edit existing images (composite, inpaint, restyle) directly, wired to the platform’s Image model. No more routing through external steps to get a concept image or a product mockup drafted mid-conversation.
Also Shipped
- Faster Copilot/Tasks sidebar: Lists load faster and stop reloading every time you reopen them.
- Steadier browser takeover: Disconnects the debug protocol during manual takeover so Google’s bot detection doesn’t flag the session.
- Signal-scan failures, surfaced: If a platform search fails during a scan, the agent now tells you why instead of silently returning zero results — and lead triage is now the agent’s job, not a manual step.
- Context usage ring: Copilot sessions show a live context-usage indicator with autocompact status, so you can see how much room is left before a session compacts.