Today’s plays › All plays › Prospecting & research
Have an agent draft every new lead's email in Slack, and learn from each rep's edits
Picked and checked by Olga Pearson, SalesUpLevel
Get tomorrow’s play in your inbox. Free ↓
Reported by LangChain (Dec 2025 to Mar 2026): lead-to-qualified-opportunity conversion up 250% and 3x more pipeline dollars in the same period. Follow-up on lower-intent leads up 97%, higher-intent leads up 18%. Reps reclaimed 40 hours a month each (1,320 hours across the team); 86% weekly active use.
Case study (the vendor's own team). LangChain's GTM agent triggers on every new Salesforce lead, first looks for reasons not to email (a fresh support ticket, a teammate who already reached out), then researches the CRM record, Gong calls, LinkedIn and the web and drops a draft in the rep's Slack with its reasoning and sources. The rep sends, edits or cancels. When a rep edits, an LLM compares the two versions and saves that rep's style notes, so the next draft starts closer to how they write.
By Vishnu Suresh and Jess Ou (LangChain), Engineer and GTM team, LangChain (vendor: LangChain; this is its own sales team) · Case study (LangChain blog) ↗ · Mar 9, 2026
If you sell: make every AI draft learn from your edits
- Trigger on every new lead in your CRM.
- Before drafting, check for reasons not to send: an open support ticket, a teammate's email this week, an active deal.
- Pull the CRM record, call transcripts and the contact's LinkedIn; if history is thin, research the company on the web.
- Write the draft from a fixed playbook that differs for customers, warm prospects and cold contacts.
- Send it to the rep in Slack with the reasoning and sources, and buttons to send, edit or cancel. Nothing goes out without approval.
- When a rep edits, save what changed as that rep's style notes and load them before every future draft.
- For lower-priority leads, set a time limit (LangChain uses 48 hours) after which an unreviewed draft sends on its own.
New lead: [name, title, company, how they came in]. Step 1, do-not-send check: here is our recent history with this account: [support tickets, emails from teammates in the last 7 days, open deals]. If any of it means we should not email now, stop and say why. Step 2, research: from [CRM notes], [call transcript highlights] and [LinkedIn/web notes], list 3 facts that matter to this person, with the source of each. Step 3, draft: write an email under 90 words for a [customer / warm prospect / cold contact] using one of those facts and one clear ask. Step 4: explain in 2 lines why you chose this angle. My style notes from past edits: [paste]. Follow them.
If you lead a team: roll it out
- Start with one lead source and keep human approval on.
- Track lead-to-qualified-opportunity rate and how often reps send drafts unedited.
Watch out for
- This is LangChain's write-up of its own team using its own product, so it is also a product story.
- Built by engineers on an agent framework with evals; most teams will start with a simpler version (CRM trigger + AI draft + Slack approval).
- The 250% covers Dec 2025 to Mar 2026 and other changes may have happened in that period.
- Auto-sending after 48 hours is a choice; keep approval required until drafts are reliably good.
Source: Vishnu Suresh and Jess Ou, 'How we built LangChain's GTM Agent', LangChain blog, Mar 9, 2026 (also summarized in The Signal, Mar 19, 2026). Case study: customer and vendor are both LangChain. Results are as reported in the vendor case study; we did not measure them.
Get tomorrow’s play before your first call
Free. One sourced play a day.