The inbox is not the product. The approval loop is.
Most social media operators I know don’t have a content problem anymore. They have a decision problem. The tools got good enough that drafting a week of posts, cutting a vertical clip, or writing five variants of a hook takes minutes. What still eats the day is the loop around the work: checking whether the client replied, confirming the flight for the shoot, reconciling what actually sold against what the affiliate dashboard claims, deciding which of forty “urgent” Slack pings deserve a human brain. The bottleneck moved from making to clearing. That’s the lens I want to use on Tend, a new invite-only assistant from maker hritvik Gupta that runs over iMessage and SMS and plugs into Gmail, calendar, Shopify, QuickBooks, and HubSpot. It is not a social scheduling tool. But the design pattern it’s betting on — text-first input, visible memory, mandatory human approval before any action fires — is exactly the pattern I think creator tooling needs to steal next.
What Tend actually is, and the problem it’s really solving
Strip away the launch-page framing and Tend is a text-message interface to an agent that can act inside your connected accounts, not just talk about them. The founder’s own diagnosis is the sharpest part of the pitch: “Most AI assistants tell you what to do—but leave you to do the work. You still have to open the app, find the email, fill the cart, or make the call.” That’s a fair hit. Ask a general chatbot to “reply to Dana that Thursday works” and you get a draft you then paste into Gmail yourself. Tend’s claim is that it does the finding and the drafting inside the account, then hands you a proposed action to approve.
The example tasks on the page are deliberately mundane and cross-functional: reply to an email, find a flight, report what sold today, find ten leads and draft outreach. For a creator or a two-person social team, that maps onto a real weekly rhythm — the brand-deal email thread, the travel for a shoot, the Shopify number you need before you post a “sold out” story, the outreach to five potential collab partners. None of it is glamorous. All of it is the tax you pay before you get to make content.
Two design principles are stated explicitly, and I want to quote them because they’re the whole thesis: “You stay in control. Nothing is sent, purchased, booked, or changed without your approval,” and “Its memory is transparent. What Tend learns about you goes into your Brain, a memory you can read and correct.” The product is invite-only during early access; you sign in at heytend.me with a phone number to request access. Pricing is not disclosed on the page. User counts, funding, and team size are also not disclosed. I’m flagging that up front because a launch post that leads with principles rather than numbers is telling you something about stage — this is early.
Why the iMessage/SMS surface is the interesting bet, not a gimmick
Every SaaS founder eventually discovers that their users live in one app they never close, and it is almost never the founder’s app. For social media managers, that app is usually Instagram or TikTok; for indie founders and growth marketers, it’s often the messaging thread with their co-founder or their client. Tend’s choice to live in iMessage/SMS means zero onboarding friction and zero tab to remember. You don’t “go to Tend.” You text it the way you’d text a competent freelancer. In my experience running lean content ops, the tools that survive are the ones that meet you in a channel you already check compulsively — which is also why so many creator workflows have quietly migrated into Slack and Notion rather than dedicated “creator OS” apps. Tend is applying that same logic to task execution.
The trade-off is real, though, and I’ll get to it in the limitations section: SMS is a thin, lossy interface for anything with structure. Approving a multi-step action over a text thread is fine. Reviewing a table of ten drafted outreach emails over a text thread is not.
How it stacks up against the incumbents you’d actually compare it to
I want to be precise here, because “AI assistant” is now a category so broad it’s meaningless. Tend sits at the intersection of three existing tool clusters, and it’s worth naming them.
The scheduling-and-publishing cluster. If your job is getting posts out the door across Buffer, Hootsuite, Later, or Metricool, Tend does not replace any of them. It has no publishing API integration for social platforms on the page, no calendar view, no analytics dashboard. What it could do, in my reading, is handle the human layer around publishing — the client asking “did that Reel go live,” the invoice follow-up for the sponsored post, the Shopify check on whether the promo code actually moved product. That’s complementary, not competitive. If you were hoping this was a Buffer killer, it isn’t, and the maker doesn’t claim it is.
The AI-agent-in-your-inbox cluster. Here the comparison is sharper. Superhuman has been pushing AI drafting and triage inside email for years; Google’s own Gemini sits inside Gmail; and a wave of “AI executive assistant” startups have tried to own the “text me and I’ll handle it” position. Tend’s differentiator, as pitched, is the combination of cross-app action (not just email) plus approval gating plus readable memory. That triple is genuinely less common than any one piece alone. Plenty of agents act; fewer show you their working memory; fewer still require a human go-ahead before every write.
The automation cluster. Zapier and Make already connect Gmail, Shopify, HubSpot, and QuickBooks. The difference is philosophical: Zapier runs rules you configure in advance; Tend runs intents you text in the moment. For a social operator, that’s the difference between “when a Shopify order comes in, post to Slack” (Zapier) and “what sold today, and draft a thank-you to the top buyer” (Tend). Both are useful. They fail in different ways — Zapier breaks silently when an API changes; an agent misreads your intent and proposes the wrong action, which is exactly why the approval step matters.
Where the approval loop earns its keep — and where it doesn’t
The comment thread on the launch is more useful than most. One commenter, Neeraj Nathany, asks the operational question I’d have asked: if a drafted reply is off, “do you edit it right in the text thread?” The maker doesn’t answer that directly on the page — not disclosed. Another, Oleksii Sekundant, zeroes in on the Brain: “A wrong assumption is much easier to deal with when you can actually find it.” That’s the right instinct. The failure mode of every memory-equipped assistant is the confident wrong assumption you can’t see. A read-and-correct memory is a trust primitive, not a feature.
And Olia Nemirovski asks the question that decides whether this is a toy or a tool for anyone with clients: “can the Brain hold a rule like never reply to this client without asking me?” Also not answered on the page. That’s the single most important open question in the whole launch, and I’ll explain why in a second.
What creators and social teams should borrow from this, regardless of whether they adopt it
Here’s where I stop reviewing the product and start extracting the pattern, because most of you reading this will not get an invite this month — it’s invite-only early access, and the integrations are B2B-flavored (QuickBooks, HubSpot) rather than creator-flavored (no CapCut, no Canva, no platform analytics APIs named). But the design moves are portable to your own workflow this week.
Move one: separate “propose” from “execute” in your own content pipeline. The reason Tend’s approval gate resonates with me is that I’ve watched too many social teams collapse those two steps and regret it. When I’ve tested AI drafting tools that publish directly, the failure isn’t usually catastrophic — it’s a wrong product name in a caption, a mistagged brand, a scheduled post that goes out during a moment of silence. The fix isn’t “don’t use AI.” It’s a mandatory review state between draft and publish. If your current stack lets an AI tool push live without a human eyeball, that’s the thing to change first, before you evaluate any new tool.
Move two: make your brand memory legible. The “Brain you can read and correct” is just a good content-ops practice dressed in product language. Every serious social operation I’ve seen has an implicit brand voice doc — the words you use, the words you ban, the client who hates emoji, the founder who insists on “we” not “I.” Most of it lives in someone’s head or a stale Google Doc. If you’re going to feed context to any AI tool — a caption generator, a repurposing tool, a scheduling assistant — that context should be a document you can open, audit, and fix. Tend’s contribution is making that memory a first-class, editable object rather than a hidden system prompt. Steal that. Write your brand Brain down. Correct it monthly.
Move three: put the boring cross-app checks on autopilot, not the creative work. The tasks Tend showcases — “what sold today,” “find me a flight,” “draft outreach” — are all low-taste, high-frequency work. That’s the correct division of labor. The mistake I see creators make with AI is pointing it at the high-taste work (the hook, the edit, the narrative arc) and hand-doing the low-taste work (the invoice chase, the collab follow-up). Flip it. Let the agent clear the inbox and the Shopify check; you keep the first three seconds of the video.
Why TikTok creators should care more than LinkedIn ones
This is a judgment call, so flag it as mine: the creators who’d get the most from a text-first agent are the ones whose income is transactional and fast-moving — the TikTok Shop affiliate who needs to know what’s converting today, the creator running a drop who needs the Shopify number before she films a “sold out” video, the UGC creator juggling ten brand threads with different deliverable deadlines. Their work is a stream of small decisions with money attached, and a text thread that can check and draft is genuinely useful. The LinkedIn creator-consultant who posts three times a week and has one client call a day? Less so. The volume of cross-app decisions just isn’t there. I’d bet the early-adopter sweet spot is commerce-adjacent creators and very small agencies, not personal-brand thought leaders.
Where my judgment says this falls short, and who it’s not for
I’ll be blunt, because a launch post won’t be.
The trust surface is enormous and the page doesn’t address it. Tend asks for access to Gmail, calendar, Shopify, QuickBooks, and HubSpot — that’s email, money, and CRM in one agent. The approval gate is a good mitigation, but the page is silent on the questions a security-conscious operator will ask immediately: Where is data stored? Is it encrypted at rest? What’s the retention policy? Can you revoke a single integration? Is there an audit log of every proposed and approved action? None of this is disclosed. For a solo creator, maybe fine. For an agency handling client Shopify and client HubSpot accounts, that’s a blocker until it’s answered. I would not connect a client’s QuickBooks to an early-access agent without a written data-handling answer.
The “rule” question is the whole ballgame and it’s unanswered. Olia’s question — can the Brain hold “never reply to this client without asking me?” — is the difference between an assistant and a liability. If Tend’s memory can only hold facts (“Dana prefers Thursday”) but not standing constraints (“never auto-send to this client”), then every task requires full manual review and the leverage evaporates. If it can hold constraints, the product becomes genuinely powerful and genuinely dangerous in equal measure. The maker didn’t answer on the page. That’s the first thing I’d test.
SMS is a ceiling, not just a surface. Texting is great for “what sold today?” It’s bad for reviewing ten drafted outreach emails, comparing three flight options with prices, or editing a long email reply. The commenter asking whether you edit drafts in-thread is asking the right question, and the silence is telling. My take: the text-first wedge is smart for acquisition, but the product will need a richer review surface (a web view, a rich iMessage card, something) for anything with structure, or it’ll cap out at trivial tasks.
Who it’s not for. If your job is purely publishing and analytics across Instagram, TikTok, YouTube, X, LinkedIn, Facebook, Threads, and Pinterest, Tend does nothing for you today — no platform integrations are named. If you’re a solo creator with one income stream and no cross-app chaos, the overhead of connecting accounts outweighs the benefit. And if you need SOC 2 or a DPA for client work, wait.
The math on “assistant” vs “workflow”
One more skepticism, because I’ve been burned: “AI assistant” products tend to demo beautifully and then hit the long tail of real accounts — the Gmail thread with twelve replies and three attachments, the Shopify store with custom apps, the HubSpot instance with 40 custom properties. Agents that work on clean demo data often stall on messy real data, and when they stall they either ask you a clarifying question (fine, but slow) or guess (bad). Tend’s approval gate protects against the guess, but the page doesn’t tell me how it handles ambiguity. That’s the thing I’d stress-test first: give it a genuinely messy task and see whether it asks or assumes.
What I’d watch / test next
If you’re curious, here’s what I’d actually do this week — no invite required for most of it.
- Request access at heytend.me with your phone number and, while you wait, write down the three tasks you’d hand over first. If you can’t name three that involve checking or drafting across two apps, this isn’t your tool yet.
- Audit your own approval loop. Pick one AI tool in your stack that can publish or send without review. Add a mandatory human step this week. That’s the Tend principle you can adopt for free.
- Write your brand Brain. One page: voice rules, banned words, client-specific constraints (“never tag this brand without approval”). This is the doc you’d feed any agent, and it’s useful even if you never touch Tend.
- When you do get in, test the constraint question first. Ask it to hold a standing rule — “never send to this client without asking” — then try to trip it. That single test tells you more than a week of demo tasks.
- Watch for the integrations roadmap. The creator-relevant integrations (platform analytics, Canva, CapCut, a social scheduler API) are the ones that would make this a social-ops tool rather than a generalist assistant. The page names none, so that’s the signal to track.
My honest read: Tend is an early, thoughtfully-designed bet on a pattern — text-first, memory-transparent, approval-gated agents — that I think is directionally right for how creators will work. The product isn’t there yet for social teams, and the trust and constraint questions are unanswered. But the design principles are worth copying today, even if the invite never arrives.
