This month we gave Agent Automations a major overhaul, opened up conversations on top of any citation, widened the model portfolio considerably, shipped seven new connectors and added TikTok as a feedback source, rebuilt onboarding end to end, made it possible to bring in your deals without turning on Sales Intelligence, and made the Agent faster on every turn. Here's what's new.
A New Experience for Agent Automations
Agent Automations launched in beta in June. This month it got a substantial overhaul — a redesigned interface for setting up, managing, and inspecting recurring Agent workflows.
- Natural-language schedules: Describe the cadence you want in plain language instead of picking from a fixed set of presets.
- Run history: Open any past run to see the whole session — transcript, tool calls, and outputs — and pick it up again in a new session.
- Everything in one place: Set the schedule, connectors, permissions, and report delivery for an automation from a single screen. Start with Enterpret Agent or build one yourself.
- Duplicate an automation: Copy a working automation rather than rebuilding it from scratch.
- Search your automations: Find a saved automation by name once you have more than a handful.
- Manage automations from a conversation: List, create, run, and update automations directly through Enterpret Agent.
- Artifacts shared automatically: HTML artifacts generated inside an automation run are now shared with recipients by default.
Every Citation Is a Starting Point
Citations used to be where an answer ended. Now they're where the next question begins.
- Add a citation directly to your current conversation, or open one and start a new chat about it.
- Attach up to five citations at once to compare evidence or investigate patterns together.
- Ask follow-ups, pivot, and dig into the full underlying data.
Every question runs against the full underlying dataset, not a copy of the rows you happened to be looking at, so you're always working with live data.
Hear the Customer, Not Just the Quote
Enterpret Agent can now create short video snippets from Gong calls. When an answer is stronger for hearing the customer directly — a deck, an account or renewal brief, an escalation, coaching material — the Agent pulls a 30–60 second clip and includes it alongside the quote.
A Much Bigger Model Portfolio
Enterpret Agent now runs on a substantially wider set of models, and admins decide which of them their teams can use.
- Anthropic's 5-series models are here. Sonnet 5 is now available in Enterpret Agent.
- GPT-5.6 Sol, Terra, and Luna joined the roster, each with six reasoning levels so you can trade depth against speed on a per-question basis.
- Refreshed open models: DeepSeek V4 Pro, Kimi K2.7 Code, GLM 5.2, and Qwen 3.7 Plus are now available.
- Model controls for admins: Choose which models and model families your teams can use, whether to meet compliance requirements or to manage cost.
Seven New Connectors
Enterpret Agent connects to more of the systems where your work already happens.
- Sentry: Rank production issues by user impact rather than raw error count, inspect stack traces and releases, run root-cause analysis, and connect a customer-reported symptom to the error behind it. Anything that changes something in Sentry asks for your approval first.
- Braintrust: Inspect projects, datasets, experiments, logs, and traces, group agent failures into patterns, and compare eval runs to spot regressions. Read-only by design.
- Mixpanel: Bring product-usage data alongside feedback to tell a small issue apart from a broad regression. Explore projects, events, and properties, run event-volume and trend queries, and inspect saved dashboards. Available for US-hosted projects today.
- Gmail: Compose and send plain-text emails from context you provide, reviewing the exact recipients, subject, and body first. Outbound-only — it cannot read or search your inbox.
- Google Calendar: List and search events, check real free/busy availability, create, reschedule and cancel events, and build a meeting brief from invite details and previous entries.
- Jira Product Discovery: Create and update ideas, search with JQL, and add customer evidence as native JPD Insights — turning feedback themes into evidence-backed ideas.
- Wispr Flow: Bring your meetings into customer context — recordings, transcripts, notes, action items, and calendar events. Turn a call into evidence and check it against the rest of your feedback, track what you committed to across calls, or prepare for a meeting using past conversations alongside the account's in-product feedback.
Enterpret Is Now Available in Canva
Enterpret is now listed as a connector in Canva AI. Ask about your customers while you're building the deck, one-pager, or campaign brief that depends on the answer, and get it back with citations that link to the underlying feedback in Enterpret.
A New Feedback Source: TikTok
Bring in comments from your own TikTok account, with captions ingested alongside them as context.
Onboarding, Rebuilt
Standing up a new workspace used to take three or four people, several steps, and significant time. Now:
- Set up a workspace in minutes, on your own. Agents do most of the work — researching the company and pre-filling the workspace description, knowledge sources, and feedback sources.
- Everything after setup runs on its own. Your taxonomy gets built and your history gets loaded without anyone driving it.
- No waiting around. Close the tab any time after setup and get a Slack or email alert when your workspace is ready.
Sales Intelligence
- Bring in your deals without turning on Sales Intelligence. Connect a deal or opportunity source and your deals come in with stage, amount, account, owner, and close date, so you can ask Wisdom things like "show me open deals." Salesforce and HubSpot are self-serve today.
- Clearer deal risk signals. Live Deal Intelligence now surfaces the specific issues that could block progress, delay a decision, or derail a deal. Set up automations to notify your team as risks emerge, so they can respond sooner.
- Salesforce write-back. Sales Intelligence can now write deal insights back to Salesforce opportunities, keeping CRM records current without manual updates.
Dashboards and Admin Controls
- HTML widgets in Dashboards: Add an HTML widget and upload a file to render an interactive artifact directly inside a dashboard.
- Metadata coverage at a glance: Unique Values and Coverage columns are now live for feedback sources under Integrations → Manage Metadata, so you can tell current fields from stale or duplicate ones.
Speed, Quality, and Reliability
The Agent got faster on every turn, and the pipelines behind it got more accurate.
- Faster answers: New questions start answering about 15% sooner. Follow-ups show progress 31% faster.
- Steadier under load: The Agent stays responsive when the system is busy.
- More accurate sentiment: A model upgrade lifted sentiment accuracy from roughly 89% to 93%.
- Better transcription: We improved our audio transcription infrastructure, so it's clearer who said what in a call.
- Sharper feedback summaries: A model upgrade cleaned up stray formatting and occasional non-English text in summaries.
- Knowledge base answers with citations: Questions against your knowledge base now come back with citations in Enterpret Agent.
- Resilient sessions: The Agent now recovers from transient errors instead of failing the session.
Bug Fixes & Product Polish
- Enterpret Agent: Forked conversations now open the right artifact, and Agent-created files are linkable and shareable end to end. Fixed errors on follow-up requests, PDF generation, model switching, and the knowledge base being unavailable in some sessions. The Agent now tells you when it stopped because of an error rather than because you stopped it.
- Slack: Enterpret Agent session and file links now unfurl with rich previews, and artifacts send correctly in Slack messages.
- Integrations: Fixed a Front integration that was only bringing in archived tickets, and restored a Salesforce Case sync that had stopped refreshing.
- Query Builder: Faster search across fields. On fields with a very large number of unique values, select the field first to find any value.
Thank you for pushing us forward. Lots more coming in August — stay tuned!
