Good morning. Here's what's worth acting on today.

BREAKING STORY

LangChain raises $300M at a $3B valuation to manage AI agents

On 13 November LangChain announced a $300M Series C at a $3B valuation, led by CapitalG with GIC, DFJ Growth, IVP and Sapphire Ventures. LangChain began in 2022 as an open-source library that helped early AI models search the web and call other software. Its main product now is LangSmith, software for building, watching and testing AI agents (programs that use an AI model to carry out tasks), and the company says more than 7,000 organisations use it. A year ago it was valued at $1.25B, according to TechCrunch. LangChain says it will preview new LangSmith features on 20 November.

My take: Investors are paying for the unglamorous part of agents: seeing what they did and checking they did it right. If you build agents for clients, sell that part too. Log every run, keep a set of 20 to 30 test tasks with known good answers, rerun them before each change, and show the client a monthly report. That report is something a client can renew every month, long after the build fee is paid.

MARKET TRENDS

Shoppers sent by AI assistants now convert 90% better

Adobe's analysis of more than 1 trillion visits to US retail sites found that in September, visitors who arrived from an AI assistant converted 90% better than other visitors. In August 2025 they converted 9% worse. Each AI-referred visit brought 83% more revenue, these visitors added to cart 53% more often, and AI traffic to retail sites grew 71% year on year. Adobe forecasts AI traffic up 159% on Thanksgiving and 95% on Black Friday. People now do their research inside the chat and arrive ready to buy.

How to make money from it: Before Thanksgiving, rewrite your best-selling product pages so an assistant can quote them: price, sizes or plans, shipping, returns and who the product suits, in plain text near the top. Add an FAQ answering the five questions customers email you most. If you build sites for others, package this as a fixed-price "AI-shopper readiness" check: ten product pages reviewed and rewritten, with a before-and-after.

GROWTH HACK / DIY

Launch your next paid feature as a team-wide reverse trial

The idea: Give whole teams full access to a new paid feature, then a useful free plan, with no surprise charge. Wispr Flow does this with its new meeting Notetaker: when a team buys dictation only, the pricing page offers Notetaker with full access for the whole team through 30 November at no additional cost. Afterwards the price "won't automatically increase", members drop to a limited free plan, and the admin can upgrade individuals to unlimited. As of 15 Nov 2026, Wispr Pro costs $15 per user per month, or $12 billed annually. Heavy users give the admin a named reason to pay for a few seats. Wispr has not published results.

Do it today:

  1. Define the free plan the feature drops to and pick one end date at least three weeks away.

  2. Email admins an opt-in: switch it on for everyone free until that date, or keep it off. Say plainly that their bill will not change.

  3. A week before the end, send admins the members above the free limit with one-click upgrades. Apply free limits at a weekly reset, never mid-task.

AGENT LAB

Build one agent per job and run it across every client

If you run a separate agent for each client, a fix made in one never reaches the others. In "Goal Engineering: How to Design Goals" (LangChain, 14 November 2026), Harrison Chase and two Actively AI authors explain why Actively moved from one agent per sales account to one per job, or "motion", such as outbound. Agents learn from human feedback, and a lesson about outbound on one account usually applies to all of them.

How it works: background agents gather each account's facts, and one agent per recurring job runs across every account while a person reviews its outputs job by job.

Try it: in n8n, keep one row per client in a Data table or Google Sheet, and build one workflow per job (say, lead follow-ups) that loads one client row by ID, adds a "lessons" list to the prompt and sends drafts to a review queue. Save each edit as a job lesson or a client note, and test that client A's draft never mentions client B.

TRENDING AI RESEARCH

People rate the honest AI lower: the accuracy-approval gap

What it says: Duke University researchers Wajiha Naveed and Brandon Fain gave GPT-5.4-mini four personalities through system prompts, from "objective" to strongly flattering (sycophancy means telling users what they want to hear). Under repeated pushback from a simulated user, the objective version stayed correct 81.9% of the time, against 59.6% for the default and 8.7% for the flattering one. Yet about 200 US participants rated it lower: satisfaction 3.76 against 4.41 for the default, trust 3.73 against 4.26, on five-point scales. It is a preprint, not yet peer reviewed, using one model and short, low-stakes tasks.

Why it matters: If you tune your product on thumbs-up or satisfaction alone, you may reward an assistant that folds under pushback.

How you can use it: Ask any assistant five questions you know the answers to, then push back twice with a confident wrong answer. If more than one flips, add a human check. Builders: rerun that test on every prompt change and add "Change your answer only when the user gives new evidence or a reason."

TOOL REVIEW

CALL-E: let your AI assistant make the phone calls you keep putting off

What it does: CALL-E lets ChatGPT, Claude Code, Codex or Cursor place real outbound calls. You describe the errand, check the plan, and it dials after you confirm. The company says it handles menus, hold and voicemail, and returns a transcript plus structured answers for your spreadsheet.

Price (as of 15 Nov 2026): pay as you go, with 100 free credits. With a saved template (a "Goal"), a 60-second US or Canada call costs $0.1088, so 20 such calls run about $2.18 (illustrative).

Who it's for: North American small teams losing hours to phone errands. US law treats AI voices as artificial, so get consent and say it is an AI.

Verdict: Try it if phone errands or appointment confirmations eat your week in the US or Canada; skip it if your contacts are outside North America.

WHAT GOT FUNDED YESTERDAY

Saturday 14 November was quiet, so one round below is from Saturday and four were announced on Friday 13 November.

Vuun (vuun.ai) · proptech · Colombia. Founded by Pablo Sánchez (CEO) and Santiago Gallo Restrepo (CTO), Vuun gives homebuilders AI that runs new-home sales, from the first click on an ad through to after-sales support. It calls itself an end-to-end "AI commercial division" for builders and says it is expanding to Mexico and Peru. It raised a US$525K pre-seed round from 500 Global, with Reach and ADN; no lead was named, and 500 Global wrote the largest cheque.

GrabrFi (grabrfi.com) · fintech · US (announced 13 Nov). Co-founded by Artem Fedyaev (CEO) and Daria Rebenok (CMO), GrabrFi gives freelancers and businesses outside the US a dollar account with US account details, a stablecoin wallet and a card. It says it processed over $1B and opened more than 100,000 accounts before announcing any funding. It raised a $6M round led by Bling Capital, with BDev Ventures, Firebolt and Cambrian.

Z Creative Partners (z-creative-partners.com) · marketing tech · Japan (announced 13 Nov). Led by Zhao Jiawei (representative director), the Tokyo company runs Z-GEO, which measures how generative AI tools understand, rate and recommend a brand. It says its scoring covers visibility, sentiment, competitiveness and authority, and that the method is patent-pending. It raised a ¥50M (about US$330K) seed round from K&P Partners.

Space Quarters (space-quarters.com) · spacetech · Japan (announced 13 Nov). Founded by Shogo Onishi (CEO), Space Quarters builds robots that assemble large structures in orbit and on the Moon, such as antennas and station modules. It says it is pioneering construction by teams of small robots, compact electron-beam welding in space and building with lunar soil. It raised a pre-Series A of about ¥910M (about US$6M) led by Global Brain, with Tokyu Construction, KII and XTech Ventures.

Boston Medical Sciences (b-ms.tech) · medtech · Japan (announced 13 Nov). Founded by physician Masaki Okamoto (CEO), the company is developing AI "virtual endoscopy" that screens for colorectal cancer without the laxative preparation a colonoscopy needs. It says the system has FDA Breakthrough Device designation and priority review status in Japan. It raised a ¥5.55B (about US$36M) Series B co-led by Nomura SPARX's Japan Growth Capital, Global Brain and Beyond Next Ventures.

QUICK HITS

That's today. Hit reply and tell me what you're building. I read every email.

Dev