Mark Zuckerberg is already shipping the broad assistant dream, OpenAI joined a day later, and Instinct raised $1 billion — a new horizontal entrant is walking into a gunfight
This week’s pick: Best bet if you have money or an audience
What you’re reading: a real Opportunity Dossier from our AI research pipeline — Ada spotted the signals, Bea shaped the candidates, Cora dug deep, Edie tried to kill it and owns the score, and Della wrote it up. New to our scoring? Start with How Our Star Score Works. Research date: October 2026 — facts move; check dates before betting money.
Mark Zuckerberg says Meta wants agents that “work day and night to help you achieve” goals. OpenAI launched Dots for paying ChatGPT users. Noah Shinn’s Instinct says more than $1 billion in transaction volume already flows through the product and just raised $1 billion more. Eugenia Kuyda, after building elsewhere, just pivoted Wabi toward “a personal agent that does stuff for you.” The category is not hypothetical anymore. The category is already a knife fight. Meta Q1 2026 prepared remarks
“Today we are starting the next major pillar of our business, Meta Enterprise Platform, to help businesses use AI to grow and transform in new ways as well.” Mark Zuckerberg said that in Meta’s September 28 launch of Meta Enterprise Platform, one day before Meta launched Muse for Small Business. Meta
“Dots are available to Pro, Business and Enterprise users, with one dot per user to start.” Axios reported that on September 29, when OpenAI stepped directly into the same lane. Axios
“More than $1 billion a year in transaction volume already flows through the platform.” Noah Shinn said that in a long-form interview about Instinct. The claim is self-reported and unverified, but the claim matters because the claim tells you what game Instinct thinks it is in: not “better chat,” but “trusted buyer.” Positive Sum
“Our personal agent that does stuff for you and builds the interface you need in the moment.” Eugenia Kuyda used that phrase when Wabi pivoted from AI app creation toward a messaging-first agent. TechCrunch
Consumer and SMB AI assistants and commerce = software that does real tasks for a person or a small business, such as answering messages, booking travel, buying products, running promotions, or moving money, instead of just chatting.
A stranger should care because the category sounds like magic but behaves like payments, customer support, trust, fraud, and distribution. An investor should care because real companies are already raising and shipping at speed. A founder should care because the market is alive but the obvious version is mostly a trap.
The broad shell is already crowded by giants with built-in distribution.
The hard part is not the chat box; the hard part is trust, approvals, payments, and ugly edge cases.
The best startup opening is one workflow, one buyer, one trust surface — not “assistant for everyone.”
The Allure
Believers are not crazy. The dream is huge.
The dream is a piece of software that stops being a passive tool and starts acting like a chief of staff. A user says, “I need to be in New York tonight,” and the software checks the calendar, finds the best flight, picks a hotel near tomorrow’s meeting, messages the colleague about a late arrival, books the ride, watches the fare, and asks for approval only at the last expensive step. For a small business, the same software could answer customer messages, pause a broken ad, reorder stock, fix a duplicate email flow, and prepare tomorrow morning’s sales summary before the owner wakes up. That is not a toy dream. That is a dream people will pay for if the software is trustworthy enough. Meta
The money story is just as seductive. Commerce is much larger than software-seat pricing. If an assistant sits in the middle of travel bookings, product purchases, ad spend, subscriptions, or vendor payments, an assistant can take a cut. Instinct is openly pursuing take-rate economics instead of advertising, and Meta already launched Muse with connectors to the Shopify catalog and later added more retail and payment partners such as Walmart, Best Buy, Shop Pay, and PayPal. Whoever becomes the trusted layer between demand and transaction could capture far more than a normal small software subscription. Meta
The installed bases are enormous. Meta said in the Meta Enterprise Platform launch that Meta already helps hundreds of millions of businesses reach customers. Meta also told investors that 9 million small businesses use at least one Meta AI ad creative tool on its platforms. OpenAI already has paying Pro, Business, and Enterprise users inside ChatGPT. A founder looking at those facts can easily talk himself into a giant future where every person and every merchant has an agent. The category-level demand is real. Meta
Believers also have one more important point: AI is not decorative here. Modern models, tool use, persistent memory, and background task execution are the core enablers. That is why so many companies moved at once in 2026. Meta
But the same wave that makes the dream possible also makes the generic version easy to copy. That is where the romance turns.
Where This Idea Came From
The spark did not come from one founder’s brainstorm. The spark came from a visible sequence of moves by real people.
Mark Zuckerberg spent months priming investors and customers for personal and business agents. In Meta’s early-2026 prepared remarks, Zuckerberg said Meta was focused on “delivering personal and business agents to billions of people around the world” and that the goal was not just an assistant, but agents that “can understand your goals and then work day and night to help you achieve them.” That was the thesis before the product wave hit. Meta Q1 2026 prepared remarks
Meta then turned the thesis into a sequence of launches. Meta launched Muse on September 8, 2026. Meta launched Meta Enterprise Platform on September 28. Meta launched Muse for Small Business on September 29. The chronology matters because Meta did not stop at the consumer dream. Meta moved immediately into business tooling, which is where startup founders often hope to hide from consumer giants. Meta
Noah Shinn lit the startup side of the fire by framing the assistant as a trusted buyer, not a chatbot. Shinn’s interview is thinly verified on performance claims, but rich on worldview. Shinn says Instinct is trying to become the user’s representative in commerce and says the company wants a merchant take rate rather than ads because an agent “smarter than its user must never be paid to persuade them.” That is a strong line, and a revealing one. Positive Sum
Noah Shinn’s own words also contain the category’s internal tension. Noah Shinn says an agent “must never be paid to persuade” the user, then describes a business model where merchants pay take rates that can be rich in categories like travel. The tension is not a gotcha about Noah Shinn’s character. The tension is inside the business model itself. If a hotel pays more than another hotel, the assistant’s incentives can drift even without banner ads. Positive Sum
Eugenia Kuyda’s Wabi pivot is another useful provenance clue because the pivot shows what smaller players are doing under pressure. Wabi did not double down on “make your own apps.” Wabi pivoted toward a messaging experience where the agent does the task and generates the interface as needed. When good founders all start bending toward the same interaction model, a market is forming. TechCrunch
The provenance story also includes infrastructure founders, not just shell builders. Natural raised $30 million in July to redesign payments for AI agents. Sapiom raised $15 million in February to help AI agents buy software tools and access application interfaces, meaning the small automatic requests one program makes to another company’s software. Those founders are telling you, with their company choices, that the hard problem is not “better prompt design.” The hard problem is operational machinery. TechCrunch
Who Wants to Build It
Plenty of founders want in. The revealing part is where founders are clustering.
Wabi pivoted in. That is a confirmed founder move. TechCrunch
Natural built agent payments infrastructure and raised $30 million. Sapiom built agent purchasing infrastructure and raised $15 million. Those are also confirmed founder moves, and those founder moves point below the consumer shell. TechCrunch
EliseAI is not a broad consumer assistant, but EliseAI matters because EliseAI shows where a real moat can form. TechCrunch reported on September 29 that EliseAI raised $350 million at a $4 billion valuation and passed $200 million in annual recurring revenue, meaning yearly subscription revenue that repeats. EliseAI got there by owning painful vertical workflows in housing and healthcare. EliseAI is the strongest “who actually wins” clue in the whole field. TechCrunch
The absence matters too. Our research did not find a wave of credible cold-start founders with no distribution trying to beat Meta, OpenAI, and Instinct head-on as a general assistant shell and already showing trustworthy traction. Most greenfield formation clusters around rails, merchant tooling, travel, household niches, and vertical operations. The founder crowd is voting with its feet.
Scorecard
Time-to-Truth — 2/5
Time-to-Truth means how fast a founder can learn whether the business is real. A trustworthy answer here takes about 6 to 12 weeks and about $2,000 to $15,000 because a founder can run landing pages, concierge tests, and 20 to 40 interviews quickly, but the real signal only appears when users hand over credentials, payment methods, or permission for real actions. Waitlist signups are weak evidence in a category already flooded with hype. TechCrunch
Time-to-MVP — 1/5
Time-to-MVP means how long to build something minimally credible, not how long to make a demo. Here, minimally credible is more like 3 to 6 months and $25,000 to $150,000 because a product needs approvals, payment or account authentication, logging, exception handling, and reliability around real-world tasks. The existence of startups like Natural and Sapiom is evidence that the “boring plumbing” is hard enough to support standalone companies. TechCrunch
Time-to-First-Customer — 1/5
Time-to-First-Customer means how fast a founder can get the first meaningful paying customer set, not a friendly tester. Expect 4 to 12 months for a broad assistant, or 2 to 6 months only if a founder narrows to one painful workflow and already has access to the buyer. Meta can pitch small businesses already inside Meta accounts and ad workflows. OpenAI can upsell people already paying for ChatGPT. A stranger starting from zero has to build trust and demand from scratch. Meta
Capital Requirements — 1/5
Capital Requirements means founder cash and total business cash needed to get credible. Founder cash is likely $10,000 to $75,000. Total year-one capital is more like $250,000 to $2,000,000 once a founder counts compute, monitoring, security, support, and one or two strong hires. That is estimate-level confidence, but even the low end is far above the usual bootstrapped software business. TechCrunch
Execution Complexity / Operational Risk — 0/5
Execution Complexity means how many ways the business can break in live use. A broad assistant that touches identities, messages, payments, third-party tools, and background actions is operationally nasty. A mistake can buy the wrong item, message the wrong person, or act with the wrong authority. Natural and Sapiom exist because these edges are not cosmetic. TechCrunch
Competitive Intensity — 0/5
Competitive Intensity means how crowded and well-funded the field already is. TechCrunch reported that Instinct raised a $1 billion Series C round at a $10 billion valuation on September 28. OpenAI launched Dots on September 29. Meta launched Muse for Small Business on September 29 after launching Muse on September 8. That is the opposite of greenfield. TechCrunch
Market Momentum — 4/5
Market Momentum means whether customers, founders, and incumbents are moving now. Momentum is real. Meta extended Muse from consumer into business use. OpenAI entered immediately. Vertical agent companies such as EliseAI show buyers will pay when the software owns a painful workflow. The problem is not demand. The problem is where the profits settle. TechCrunch
AI Leverage — 4/5
AI Leverage means whether AI creates real new capability or is just decoration. AI is the core enabler here. Action-taking agents, memory, tool use, and background execution are the category. The warning is that the same AI also commoditizes thin wrappers, so strong technical novelty does not guarantee a moat. Meta
Defensibility — 1/5
Defensibility means whether a business can stay valuable after copycats arrive. A standalone assistant shell is weakly defensible unless the company captures rare trust, proprietary workflow data, or embedded distribution. Meta already has distribution across Instagram, Facebook, WhatsApp, and ad accounts. OpenAI already has distribution inside ChatGPT. The realistic moat is vertical workflow ownership, not “we also have an assistant.” Meta
Key economics
Your money: founder cash $10,000 to $75,000 versus total capital $250,000 to $2,000,000 in year one. Both figures are estimates, not audited base rates. The gap matters. A founder may be able to start tinkering alone, but a founder cannot seriously contest the broad shell with tinkering money. TechCrunch
Time to break-even: 12 to 36 months, estimate and speculative, assuming a founder narrows to one recurring SMB workflow with subscription or take-rate revenue instead of trying to monetize a broad free assistant. Without that narrow wedge, we could not model break-even honestly. TechCrunch’s reporting on ugly consumer AI economics points the same way. TechCrunch
The audience you’d need first: about 2,000 already-reachable niche operators or merchants who share the same workflow pain and already trust the founder enough to try automation. “Already reachable” means people the founder already knows through past work, an agency, a customer list, or a real operator reputation. Without that connection, add roughly 12 to 24 months of trust-building and sales effort before the same business has a chance. Meta
Cost per completed outcome: for the first pilot customer acquired by founder-led direct outreach in a tight niche, expect roughly $100 to $800 in founder time and tooling per pilot customer. For broad paid acquisition of a trust-heavy assistant, cost per completed outcome is not honest to model yet because click costs may be measurable while successful credential-sharing and retained transaction behavior are too variant and too weakly evidenced. TechCrunch
What moves the needle: for a cold stranger, this stays 1 star. With a real niche audience, the opportunity rises to 2 stars because the founder may reach pilots fast enough to discover a narrow wedge. With capital, the opportunity also rises to 2 stars because reliability and integrations improve, but weak moat remains. With both money and audience, the opportunity reaches 3 stars only if the founder avoids the generic shell and becomes the vertical workflow or rails layer. No amount of money or audience fixes the structural ceiling of a weakly defensible new horizontal shell against incumbents with native distribution.
The person who wins here is: someone who already lives inside one messy commerce or SMB workflow, knows the ugly edge cases from firsthand pain, has warm access to buyers, and is patient about trust, approvals, and operational cleanup. That person is happy to be the boring but critical layer.
The person who’ll be tempted but shouldn’t be is: the smart solo builder dazzled by viral demos, the marketer who thinks audience alone can rescue a me-too assistant, or the operator who wants the AI wave but hates technical ambiguity and ongoing failure handling.
What It Would Take to Raise This Score
A founder would need to stop pitching “assistant for consumers and SMBs” and instead prove one narrow workflow with real authority. A founder would need 20 to 30 interviews with one buyer type, 5 to 10 concierge pilots that require a true trust step, and at least 3 of 10 target users willing to connect an account, share a payment method, or approve a real action within 21 days. A founder would also need evidence of repeat use around one recurring job and a realistic gross margin after model-inference cost, payment cost, monitoring cost, and human rescue cost. Without that proof, the score does not move honestly.
Go-to-Market
The best channels for this variant are partnerships, cold outreach / direct messages, and marketplace or directory distribution.
Partnerships — strong fit. The product needs borrowed trust. The smartest version plugs into an existing ecosystem such as Shopify apps, agency partners, bookkeeping tools, or a platform where the buyer already has accounts and habits. Cost and effort: usually 10 to 30 partner conversations for the first serious channel, plus integration work. Customer acquisition cost per pilot can be lower than cold paid media because the platform or partner carries the trust burden.
Cold outreach / direct messages — strong fit. Cold outreach works only when the product solves one named workflow pain for one named buyer, such as travel-heavy service businesses, property managers, or merchants with duplicate message and ad operations. Cost and effort: 50 to 150 highly targeted outreach notes for the first 5 to 10 pilot conversations; roughly $100 to $800 in founder time and tooling per pilot customer.
Marketplace / directory distribution — strong fit. Listing inside the place where intent already exists is far better than teaching the world why a new assistant matters. Shopify app stores, plugin directories, and integration marketplaces are plausible fits if the wedge is narrow enough. Cost and effort: integration plus listing prep, often 2 to 6 weeks after a real wedge exists.
First 90 days should look plain, almost boring. Weeks 1 to 3: pick one workflow and one buyer, then run 20 to 30 interviews. Weeks 4 to 6: do 5 to 10 concierge pilots manually or with lightweight automation. Weeks 7 to 10: ship only the approval, authentication, and reporting loop needed for that one workflow. Weeks 11 to 13: turn the best pilot results into case studies and partner pitches.
One thing a day: send one outreach note, run one user interview, review one failed workflow, publish one tiny proof artifact, ask one partner for one introduction, tighten one approval step, and follow up with one pilot every day.
Considered and set aside: paid social, influencer, press and Help a Reporter Out, launch platforms, awards, local offline channels, broad paid search, generic organic social, own podcast or YouTube, affiliate programs.
Illustrative case study — not a real company
A former agency operator knows 3,000 Shopify merchants and keeps hearing the same complaint: post-sale support is scattered across Instagram messages, email, and order tools, and holiday rushes create duplicate replies, missed refund deadlines, and angry customers. The founder does not build a “personal commerce assistant.” The founder offers one promise: “approve, send, and log customer-service resolutions across three channels from one queue.”
In week one, the founder interviews 12 merchants the founder already knows. In week three, 6 merchants agree to concierge pilots. Three merchants allow inbox access. Two merchants allow limited refund approvals under dollar thresholds. One merchant refuses any account connection and drops out. By day 21, only 2 of the 6 merchants have crossed the trust threshold enough to let the workflow run repeatedly. That sounds small. For this category, that is valuable truth.
The founder then discovers the real product is not drafting replies. The real product is approval routing, refund rules, and audit logs. A flashy horizontal demo dies. A boring workflow tool lives. That is what a salvage path looks like here.
The honest skeptic’s case
The skeptic’s case is brutal and mostly right.
Meta can already wire an assistant into surfaces millions of businesses use. OpenAI can roll a competing assistant directly into paying ChatGPT accounts. Instinct just raised a billion dollars. A new founder trying to launch “our own AI assistant for consumers and SMBs that also handles commerce” is choosing the most crowded and least defensible layer while also choosing one of the worst support burdens in software. TechCrunch
The economics are also ugly. Consumer assistant use can require persistent context, repeated tool calls, payment checks, retries, background runs, and human rescue. TechCrunch’s late-September piece on consumer AI economics is a useful reality check: usage may look magical long before margins look sane. TechCrunch
Trust can break fast. TechCrunch reported backlash to Instinct’s recommendation feature, with some users saying the suggestions gave them “the ick.” That phrase sounds trivial until a founder remembers what kind of product this is. In a trusted assistant, a creepy suggestion is not a minor user-interface annoyance. A creepy suggestion is evidence that the product may be acting for itself, not for the user. TechCrunch
Some of the hottest category numbers are also shaky. Instinct’s transaction volume, retention, credit-card sharing, daily growth, and zero-marketing claims are self-reported and unverified in our research. A founder who builds a company plan on top of those numbers should weight them down sharply. Positive Sum
The thesis-killer alternative
The thesis-killer alternative is simple: build on top of the winning assistants and rails instead of trying to become one.
Why build a new general shell when Meta has Muse, OpenAI has Dots, and infrastructure startups already handle payments and purchasing for agents? A founder can dodge the hardest requirement — earning general-purpose trust plus distribution — by building a vertical agent, approvals layer, merchant optimization tool, or identity and payment rail on top of existing ecosystems. That bypass is strong enough that, for most founders, the bypass does defeat the broad opportunity. Axios
The bypass does not kill the whole category. The bypass kills the naive version. A founder can still win by becoming the best layer for one vertical job, one merchant stack, or one risky approval flow.
Business model, MVP, time-to-truth test, risks, reasons not to pursue
Business model: the honest early models are subscription for a narrow SMB workflow, usage pricing tied to completed tasks, or a take rate on transactions in a category where commissions are already normal. Travel is the obvious example, but travel also comes with fierce competition and mature incumbents.
MVP: not a general assistant. The MVP is one workflow with one approval pattern and one report the buyer cares about.
Time-to-truth test: within 21 days, recruit 10 narrowly defined target users and ask the 10 target users to complete one real delegated commerce or operations task through a concierge prototype that requires a meaningful trust step. Meaningful trust step means connect an account, share a payment method, or approve a real purchase or message flow. If fewer than 3 of 10 target users cross that threshold, or if the workflow requires heavy human rescue, kill the horizontal assistant idea immediately.
Main risks: distribution risk, trust risk, gross-margin risk from model and tool cost, platform dependency risk, support risk, safety risk, and “looks magical in demo, collapses in live use” risk.
Reasons not to pursue: no niche audience, no direct buyer access, no appetite for support and exceptions, no willingness to sell before building, or a thesis that still sounds like “better assistant for everyone.”
The full top 3 takeaways
The broad shell is already crowded by giants with built-in distribution. Meta launched Muse on September 8, then moved into Meta Enterprise Platform on September 28 and Muse for Small Business on September 29. OpenAI launched Dots on September 29. TechCrunch reported Instinct’s $1 billion round at a $10 billion valuation on September 28. A new entrant is not walking into a blank market. A new entrant is walking into a market where the top of the stack is closing fast. TechCrunch
The hard part is not the chat box; the hard part is trust, approvals, payments, and ugly edge cases. Natural and Sapiom are real companies with real funding because agents need ways to move money, buy tools, authenticate, and recover from failure. The operational mess is the product. A founder who mistakes the shell for the substance will learn expensive lessons slowly. TechCrunch
The best startup opening is one workflow, one buyer, one trust surface — not “assistant for everyone.” EliseAI is the best proof point in adjacent territory: vertical workflow ownership can become enormous. The opening for a new founder is not “beat Meta.” The opening is “remove one painful loop inside a stack a buyer already uses, then become the trusted approval and execution layer there.” The category is real. The broad idea is weak. The wedge idea is alive. TechCrunch
The unresolved question — the question that will decide the whole category — is whether trust compounds faster than distribution. If users really will hand one company the calendar, the inbox, the card, the ad account, and the right to act, an independent winner can still emerge. If distribution wins first, then the big platforms own the shell and startups get pushed into the rails and the verticals. Nobody knows that answer yet, and that uncertainty is the whole story.
Provenance: Super Fluidity Opportunity Dossier, public method format, AI-assisted analysis, facts source-verified where possible and labeled cautiously where single-source or self-reported. Corroborated facts were checked against at least two independent sources when available; self-reported company metrics were not treated as confirmed proof.
Why five ratings? Because the same idea is a different bet for different people. As a side gig — no money, no audience, no special skill, a few hours a week beside a job — the question is simple: is there a realistic path to $2,000–$5,000 a month within a year? As a business, starting with no money and no audience, most ideas are traps — even ideas that became famous companies rate only 2 stars from that starting line. With money, an audience, or both, the same idea can be a real opportunity. So every dossier rates all five, and you can find your own row.
Dossiers rated 0–2 stars for a founder with no money or audience are published free, right here. The rarer 3-, 4-, and 5-star finds go to paid subscribers first — that tier is opening soon, and free subscribers get the first invitation. Join the free list. Or have us investigate your own idea, privately: Request a Dossier ($199).
