Nathan Weill here.
I'm the CEO @ Flow Digital, where we help companies unleash their full potential by strategically automating every inch of their workflows. Each week, I share hot AI and automation tips to help you move your business into the future successfully.
Tend: Learns Your Messages, Drafts Replies
Triage email, Slack, texts, and calendar
Operations leaders lose hours a day to the same loop: read a message, decide what to do, act, and repeat, across email, Slack, calendar, and docs. The cost is decision fatigue and dropped balls, where a client reply or approval stalls before it reaches the top of the pile.
Tend, a new experimental, open-source release from Every, collapses that loop: it does the reading and the first pass of the thinking, then sharpens every time you use it.
Every item becomes a decision-ready card: Tend converts each message across Gmail, Slack, Google Calendar, Notion, and iMessage into a card with a plain-English summary, a drafted reply, or a suggested next action, so you decide in seconds.
It learns from every call you make: each time you approve, edit, skip, or redirect a card, Tend absorbs what matters to you, so its summaries and drafts get more accurate and each sweep asks less of you.
Nothing sends without your sign-off: you stay the approver on every card, so the agent drafts and proposes but never acts on its own.
Tend launched this week as an experimental project, so broad independent reviews are still thin. The early signal:
What's working:
Decision-ready cards save the most time: early reviewers point to the summary-plus-draft format as the feature that turns triage into quick approvals.
It gets better with use: the learning step after each run is singled out as the difference between a one-off assistant and a system that compounds.
Human control by default: the approve-before-send design is the headline safeguard and works as intended.
Limitations:
Setup is hands-on: you copy a prompt into Codex or Claude and connect your own sources, which takes setup effort and technical comfort.
Weaker on design-heavy work: reviewers note the model handles text well but is less reliable on visual or intricate output.
Not for critical workflows: Every ships it with no support, no stability guarantees, and expected breaking changes, so keep it clear of irreversible actions.
HubSpot
Automatic stakeholder and pain-point signals
HubSpot Intent Signals
The two facts that decide a deal, who signs off and what problem they need solved, usually sit buried in call transcripts and email threads nobody reopens.
HubSpot's two new intent signals read that engagement for you. Key Stakeholder Identified fires when a decision maker, champion, economic buyer, or influencer appears in a conversation, and Pain Point Mention fires when a contact names a business challenge. Both are extracted from your deal-associated emails, notes, and call transcripts.
Detect the decision layer automatically: know the moment a champion or economic buyer enters a deal, not three calls later.
Log stated pain as data: the exact problem a prospect named becomes a signal, not a line lost in a transcript.
Act on both anywhere: each signal is eligible across Workflows, Scoring, and List Segmentation.
What you get:
Qualify faster: re-score or route a deal the moment a real buyer or a clear pain surfaces.
Personalize follow-up at scale: trigger outreach that references the exact pain the prospect stated.
Protect the base: a fresh pain point on a current account cues a proactive save before it turns into churn.
You already capture this on every call. These signals finally make HubSpot act on it. See HubSpot's intent signals overview and its buyer intent tools.
Communicate Design Ideas
Stop asking your human or AI designers to move “the thingy”
Another tool recommendation from the tool nerds at Flow Digital. This one is small, but so useful, it had to be included.
Describing what you want built is where design ideas stall. You say "the three-dots menu" or "that gray text that disappears," the developer or agency guesses, and you lose a round of revisions to a vocabulary gap. With AI coding agents now in the mix, a vague word produces the wrong screen just as fast.
NameThatUI is a free visual dictionary of UI web design terms that gives every element its real name. You find the component, learn what it is actually called, and hand off a term the builder, or the AI, cannot misread. Its pitch: "See the element, learn its real name, and prompt your coding agent with precision."
Search in plain language: press ⌘K, describe what you see, and get the precise term.
Look it up visually: browse a catalog of Web and macOS elements, and double-click any word for a plain-English definition.
Prompt with precision: paste the correct name into your brief or AI coding tool so it builds the right thing the first time.
Shared vocabulary removes the back-and-forth between the person with the idea and whoever, or whatever, builds it. That means fewer wrong builds and faster turnarounds on every site, app, or generated screen, and it costs nothing to start.
Automating Customer Data to Protect Brand Voice
AI is flattening brand voice. Your own customer data is the way out.
Your brand is starting to read like your competitors', and AI is quietly the reason. The State of Brand calls it a "Great Flattening": feed an AI model the same adjectives every rival uses, and it returns the same voice.
Two numbers frame the stakes. Ahrefs found 74% of new webpages already contain AI-generated content, the very corpus these tools learn from.
Why Sameness Costs You
Customers notice. Klaviyo's 2026 report, covering 8,000 consumers across eight countries, found only 13% fully trust AI, while 31% trust a brand less when they spot AI in its marketing, against just 7% who trust it more. Even self-described AI enthusiasts are wary: 39% would trust a brand less for AI-written content.
The fix: build voice from evidence, not adjectives
The article's remedy is to ground voice in real beliefs, decisions, and positions a competitor could not copy. Our addition: the raw material for that already sits in your systems, and you can automate the sourcing so it stays current. Instead of describing how you want to sound, mine how your customers actually talk and what they actually need.
Run a four-source data sweep
Point automations at the places customers reveal themselves, and refresh on a schedule:
Support and FAQs: pull helpdesk tickets and chat logs (Zendesk, Intercom, or your HubSpot inbox) into a weekly digest that ranks the real questions and the exact phrasing customers use.
Sales conversations: run call transcripts through conversation intelligence to surface stated pain points and objections, then bank the verbatim language for messaging.
Revenue versus attention: cross-reference CRM sales data against your content calendar to see which products and services actually drive revenue versus which ones marketing over-promotes.
Reviews and surveys: auto-collect review, NPS, and survey verbatims onto one board that flags the outcomes, objections, and beliefs customers repeat.
Find the talk-versus-reality gap
This is where most brands are hiding their differentiation. Tag every customer input by theme (product, use case, objection, outcome), then compare two frequencies: how often a theme shows up in customer data versus how often it shows up in your own marketing. Two lists fall out:
Over-indexed: themes you talk about far more than customers do, usually internal pride rather than a buying reason.
Under-served: pains and outcomes customers raise constantly that your messaging barely mentions. This is your voice, unclaimed.
Turn the sweep into a voice document
Feed the AI fuel, not adjectives. Replace the adjective list with a living document that captures:
Beliefs and positions: what you argue that rivals would not, including the trade-offs you make and the customers you decline.
Customer language bank: the actual words, questions, and phrases from your data, grouped by theme.
Proof and specifics: numbers, names, and stories only your company can tell.
A never-say list: the clichés and tics to strip out, starting with the "not just X, it's Y" cadence that AlphaSense found in 73 corporate filings in a single quarter, after barely appearing for two decades.
Where humans stay in the loop
Automation surfaces the patterns; people make the calls. A human decides which beliefs the brand will stand behind, checks that customer language is represented fairly rather than just loudly, and edits every draft for truth. Route transcripts and tickets through privacy review first, since they carry personal data. The machine finds the signal; your team gives it a spine.
Watch-outs
Don't mistake volume for truth: the loudest customers are not the average customer, so weight by segment and account value.
Don't mirror complaints back as voice: pain language guides tone and topic, it does not become your tagline.
Don't set and forget: re-run the sweep quarterly, because both customer language and your own drift over time.
Go deeper
Read The State of Brand's original argument and its data-backed follow-up for the full research trail.
Build Your Own Voice Automation
This is the kind of workflow we build. Book a free 20-minute Discovery Session and we'll scope a voice-sourcing automation for your stack
SMART WORDS OF THE WEEK:
— Dan Shipper, Every
Our Take: The teams that pull ahead this year will aim their automation at the right decisions, then stay in the loop to make them.
 |
Stay in touch with our team from anywhere.
