We scale AI-powered customer conversations—human-like in depth, machine-like in scale—to deliver the clarity you need to grow. #AI#CustomerClarity#Growth
1/
Everyone's trying to ship faster. Doesn't matter if you're a veteran PM or solo vibe coder, validating before you build has always been a bottleneck...until now.
Watch us create a voice agent that interviews users to collect feature requests in 30 seconds.
Next: Embed in-app
🚀 Latest release
🆕 When you review a completed conversation, you now see the same closing message your participant saw at the end, including the next step you guided them to. Reviewing a conversation now reflects how it actually wrapped up for each participant.
⚡ On the analysis page, the suggested prompts now sit just below the chat input and tuck away the moment you start typing, so the page reads cleanly from the input down to your session history without jumping.
🐛 When your closing instructions ask a conversation to end on a specific final line, the agent now follows them more faithfully. In rare cases it could add its own wrap-up or a second goodbye after your closing line; conversations like these now close just as you wrote them.
🐛 Once a conversation is complete, reopening it can no longer slip in a stray message or trigger another response from the agent.
🐛 When Perspective is removed from a Slack channel, that channel no longer lingers as a dead option in the channel picker, so the channels you can route conversation outcomes to are always ones that still work.
https://t.co/pNSoLRrR52
🚀 Latest release
🆕 Analysis now works on preview conversations, not just live ones. Since previews are saved alongside the rest of your conversations, every conversation you collect, including the ones you ran while testing, can feed your analysis instead of being left out.
⚡ Connecting Slack on a large workspace now reliably finds every channel. The picker used to show a channel count that changed between refreshes and leave out channels you had already invited Perspective into. It now loads your full channel list in the background, keeps it current as channels are added or archived, and lets you add a channel by its ID when search comes up empty, so conversation outcomes route to the Slack channel you actually chose.
🐛 Voice conversations now reliably end the way a call should. In rare cases a call could linger after the agent's closing line, repeating its goodbye instead of wrapping up; now it closes promptly every time, with your transcript and recording saved before the page moves on.
🐛 Voice conversation transcripts now reliably read in the order things were said, with timestamps that match when each person actually spoke. In rare cases a participant's answer could land out of order, after the reply it prompted; reading back a call now always follows the real flow of the conversation.
🐛 Demo conversations you generate now finish with a proper closing message instead of cutting off on the participant's last reply, so they read like a real conversation from start to finish.
https://t.co/lkYYkv5nQ2
🚀 Latest release
⚡ Ask the design agent to review your conversation outline and it now evaluates it across multiple dimensions: whether it captures what you set out to learn, whether it follows solid conversation-design practices, and whether it's cleanly structured. You get a score on each, so you can see exactly where the outline is strong and where it needs work, plus specific numbered suggestions you can apply all at once or pick from, like "apply 1 and 3."
🐛 Automations that send participant emails through a connected Gmail account are now recorded correctly. They were delivering fine but getting logged as failed runs, which made working automations look broken in your history.
https://t.co/L0gpM0yKcU
🚀 Latest release
🐛 Conversations now open the way a person would: a warm greeting and the first question, instead of an out-of-place reaction before you've said a word. Participants feel welcomed from the very first line.
⚡ Embedding a voice conversation in a page builder like Wix or Squarespace? New troubleshooting guidance walks you through fixing the "microphone access blocked" error these wrappers can cause.
https://t.co/qnO5kEUlEK
🚀 Latest release
🆕 When you connect an AI assistant to Perspective, like Claude, ChatGPT, Cursor, or another MCP client, you can now choose which workspaces it can access rather than granting every workspace you belong to. Pick the workspaces when you approve the assistant, and see exactly where each connected app has access from your settings.
⚡ Gemini 3.5 Flash is now featured as a recommended model in the research AI chat, so the faster, newer option is front and center when you pick a model.
https://t.co/lDe6Eg6OV5
🚀 Latest release
🆕 Conversations feel more natural and human across text, voice, and phone, with a smoother experience from start to finish.
🐛 Previewing a conversation no longer shows a mismatched header color. Previews look exactly like the live view, with the preview label as the only difference.
https://t.co/6Ysw28mdNM
🚀 Latest release
⚡ Perspective now reasons through your goal and plans the structure before writing your Outline, so it captures your intent more accurately and follows best practices for asking good questions. You can watch that thinking unfold live as it works.
⚡ Generated Outlines need less cleanup: each section now stays distinct instead of repeating or blending into others the way it occasionally did, so what you get is closer to ready to use.
🐛 Your Outline's versions now stay accurate: rebuilding from scratch continues the version count instead of resetting to Version 1, and an edit that changes nothing no longer adds an empty version.
https://t.co/6ilZWrU0Pd
🚀 Latest release
⚡ The MCP keys settings screen is redesigned around your keys. When you create a key, its token now appears right away in a copyable box so you can save it before closing the dialog, you pick which workspaces a key covers in one step, and the steps to connect Claude, Claude Code, or Cursor are grouped together in one place.
https://t.co/aLoQyHAqzJ
🚀 Latest release
🆕 Analysis sessions can now chart your conversation data as line, pie, stacked bar, scatter, and heatmap visualizations, not just bar charts and tables, so trends, proportions, and outliers are easy to spot at a glance.
🆕 You can now create multiple MCP keys and scope each one to specific workspaces, replacing the single all-access token. Connect Claude, Claude Code, or Cursor to only the workspaces a key allows, and revoke keys individually without disrupting your other connections.
⚡ When you make a copy of an agent, you can now keep its original owner instead of taking ownership yourself. You're added to the copy as a member, so you keep access either way.
https://t.co/2CKbcJSh9h
🚀 Latest release
⚡ The primary button in your research sidebar now reads New agent instead of New conversation, making it clearer that you're spinning up a new AI agent.
https://t.co/KBB4xgIWen
🚀 Latest release
🆕 Have a specific Perspective question? Every docs page now has an Ask Perspective button. Click it to ask the Perspective assistant directly and get a guided answer drawn from our docs, so you don't have to read through them to find what you need.
🆕 The Perspective SDK now supports a "push" mode for the slider embed. Instead of overlaying the page, the slider takes its own column and lets the rest of the page stay fully interactive.
⚡ Perspective's marketing site, docs, and blog are now properly discoverable to AI assistants and crawlers. Every page advertises pointers to our machine-readable resources, and blog and docs pages expose a clean text version of their content so agents do not have to parse HTML.
https://t.co/MKirEBu3om
🚀 Latest release
🐛 Signing up no longer breaks if the welcome interview captures a name with unusual characters. Your name is cleaned up automatically and you land in your workspace instead of getting bounced back to sign-in.
https://t.co/CntNpPjjhv
🚀 Latest release
⚡ Your workspace view is now a clear list of all your agents (Interviewer, Concierge, Evaluator, and Advocate), each showing its type and whether it's a draft, ready, or active, with a new filter to narrow the list by agent type.
https://t.co/gzIp6V8BSe