InstaAgent: YC's 19th bet on AI ad generation.

Kyle Wong (Co-Founder & CEO) and Colin Tseung (Co-Founder) have known each other since middle school in Hong Kong.

Both are repeat founders. Colin previously ran Cyclopes Career from 2022 to 2023 and continues to list a part-time role at GuildEra, beginning in 2024. Kyle's public record shows a nine-year run at Goldman Sachs, from Analyst to Executive Director in China TMT. Founder bio claim (1M+ users in one month; a prior acqui-hire) names no project or acquirer. Undisclosed acquirers and small unlisted projects are ordinary, so absence from public sources is expected rather than notable.

Neither founder's public employment history shows a dedicated role in paid social, creative production, or agency operations, though prior founder roles would not necessarily surface one.

Against its three same-batch category peers, InstaAgent leads on team size and launch votes and trails on LinkedIn reach and listed skills. The set is technically credentialed: CharacterQuilt, Kuli and Memoir bring research and engineering backgrounds from NVIDIA, EPFL, Stripe and IMC Trading. InstaAgent's founders bring finance and enterprise-network access instead, which is a different bet on what wins this category rather than a weaker one, and the seven staff this dataset cannot see may already cover the gap.

InstaAgent positions itself as taking a single creative brief and producing hundreds of persona-tailored ad variants for Meta and TikTok, then using live performance data to kill underperformers and scale winners on a continuous loop. Its business model and mechanics are more precisely defined in their May 2026 launch post: the company started as a self-serve AI tool and switched to an agency model after learning that "customers do not just want AI tools; they need help producing quality creatives".

Self-reported figures are not thereby suspect; they are simply unaudited, as everyone's are at this stage.

The fulcrum of this analysis is the $2M ARR figure, and how it is counted matters as much as its size. At $2M across more than 500 clients, implied average revenue is roughly $4,000 per client annually, or $333 per month. That figure is a ceiling rather than an estimate: "500+" is a floor, so a larger real client count would push the implied average lower still.

$333 per month sits below what a fully managed creative service would typically charge corporate accounts, so the average is probably not describing a uniform roster. Three ordinary explanations could fit; public data does not distinguish among them. The denominator may be broader than it reads, if "clients" counts trials and one-off engagements or accumulates relationships rather than tracking current ones. The average may simply be unrepresentative, with revenue concentrated in a handful of larger accounts above a long tail of smaller ones. Or the numerator may include work beyond recurring service fees. All three are normal at this stage; without pricing, concentration, or revenue-recognition detail, public data cannot tell which applies.

Icon offers the nearest published price point at $1,000 per month, three times the implied figure, but the comparison is directional at best: Icon has repositioned to explicitly human-made UGC, six human-produced ads monthly, so its floor is set by the cost of paying people to film. A model built on AI-generated variants should cost less per unit. Cheaper is the expected result, not the anomaly. What is harder to place is how far below.

Separately, revenue composition affects quality as well as size. If any portion of the $2M represents pass-through media spend rather than net service revenue, the economic value of the reported top line would be materially lower, and the implied service revenue per client lower again. There is no public evidence that this is the case; it is a diligence question, not a finding.

The market InstaAgent sells into is worth setting out on its own terms. Four things can be measured: how much budget marketers hold, how much is moving to AI, what the platforms carrying it are building themselves, and how many companies have arrived to split the rest. They do not point the same way. The force creating the opportunity is the one crowding the category too.

The buyer's budget is not collapsing, it just stopped moving. Gartner puts marketing at 7.7% of company revenue in 2024 and 2025 and 7.8% in 2026, well below the ~11% that held pre-pandemic. A flat pool is the setting for everything below. 

Inside it, AI is where the money moves: CMOs allocate 15.3% of marketing budgets to AI, 21.3% among organisations Gartner rates AI-ready, while only 30% report mature AI capability. Real budget unevenly matched by the ability to spend it is exactly the gap a managed service fills. Aggregator sizings agree on direction: gen-AI in advertising $3.37B to $4.18B on a ~24% CAGR toward $9.81B by 2030, the broader marketing definition $4.89B to $6.58B toward $18.29B. 

Consumer sentiment is also evolving. Pew Research found 50% of Americans feel more concerned than excited about AI's role in daily life. Disclosure carries a measurable cost: field studies find AI-labeled posts draw ~7-8% fewer likes (Journal of Consumer Research, 2026) and up to 14 points less behavioral engagement (Electronic Markets, 2026). For advertising specifically, disclosing genAI use cut click-through rates by 31.5% in a large field experiment (NYU Stern, 2025). For a company whose pitch is 'we'll make you more ads, faster,' this is a headwind.

AI visibility is Semrush's 0-100 score for how often a brand appears in AI-generated answers relative to competitors. 'Mentions' counts the distinct prompts in which it surfaced while 'Cited pages' counts how often the company's own pages were used as a source. The four assistants covered are ChatGPT, Google's AI Overview and AI Mode, and Gemini. This channel matters because this is where consideration sets are now assembled. A marketer asking an assistant which tools generate ad creative receives a shortlist, and companies outside it are never evaluated. Fifteen of the nineteen register at all. InstaAgent surfaces in three answers, two in ChatGPT, one in Google's AI Overview, none in Gemini or AI Mode, and registers AI presence in a single market, the United States, against Variant's 21 and Lapis's 13. 

Additionally, every one of the ten companies with more mentions has at least one cited page, from Bluma's 6 to Stormy's 485. InstaAgent, on the other hand, has no page of its own being used as a source. Assistants discuss it entirely through other people's pages: Y Combinator's profile, Reddit, LinkedIn. 

The visibility score, however, is a floor, and not a ranking. 14.0 is shared by ten of the fifteen companies registering at all, so InstaAgent sitting at the industry median reflects the metric's entry value rather than a position earned. The leaderboard is also not an accurate depiction, the top companies' cited sources point substantially at unrelated businesses sharing their names, which means the gap between InstaAgent and the field is narrower than the raw count suggests.

YC has funded eighteen other companies building AI-driven ad or marketing creative, spanning Summer 2021 to Spring 2026, making InstaAgent the nineteenth. Spring 2026 contributes five, Fall 2025 five. InstaAgent is entering an established and increasingly crowded category rather than creating a new one, and sits 16th of 19 by the one yardstick all of them share.

Priced in dollars the gap is starker than the rank. Semrush values InstaAgent's 92 visits at $16 (as of June) of equivalent search advertising, Plai's 4,231 at $25,570 (as of June), Variant's 18,017 at $19,100 (as of June). Measured in traffic value rather than rank, the gap to the category leaders is roughly three orders of magnitude.

Headcount sharpens it: nine staff is the fourth-largest team in the set, behind Plai (20), Uplane (19) and Affogato (15), while Kuli produces 2,873 visits with a team of one. Whatever the nine-person team is building or delivering, it does not yet show up in search traffic. It may show up in client delivery instead, which no search tool measures. Plai shows the clearest evidence of scale in the peer set. Variant leads on traffic, but because "variant" is a common English word, some portion of its 18,017 visits may be non-branded or unrelated; its headline total should be treated cautiously. This finding is also supported with more recent data, stating that Plai is now the leader by traffic value (3,059 traffic against $24,075 traffic value) while Variant leads by volume (25,700 traffic against $17,609 traffic value).

Plai: 5 years old, 20 employees, 100K+ campaigns, 5B+ impressions. The clearest evidence of scale in the peer set. Lapis: 1,000+ teams including Hyundai and Tata, ranks #3 for “ai ad,” early mover on ChatGPT Ads. Variant is the set's most opaque traffic leader: 25,700 estimated visits, but little public disclosure and a name collision that makes the headline number hard to interpret. Two prior founder exits. Hiring at $110K-$300K.

InstaAgent covers Meta and TikTok: two platforms. Plai covers 10+. Lapis has added ChatGPT Ads. For brands seeking multi-platform reach, this is a constraint. The focused approach can be a strength, but it creates dependency on two platform ecosystems that are actively building competing native tools.

Public tools show no tracked paid-search activity for InstaAgent. That does not establish whether the company advertises on Meta or TikTok, whose campaign data is not comparably visible. Authority score: 17 out of 100. Organic keywords: 53, and of the seven ranked terms visible, six are misspellings of the company's own name. Paid keywords: zero. The top ranked keyword is "instaag" at position 11, a misspelling of the company's own name, that appears on page 2. Nothing the company ranks for appears on page one of Google for anyone. That traffic is worth roughly $15/month in equivalent Google Ads spend. Plai's is worth $24,075. The YC P26 batch average sits around ~500 monthly visits and an authority score of 6.5. InstaAgent sits well below the batch average on traffic and well above it on authority.

Two benchmarks, two different questions. Their traffic plateaued at 98, against a median of 205 across the nineteen companies in its category. Against its own Spring 2026 batch, InstaAgent beats 79% to 92% of the field on all five metrics. Against the 3,035 YC companies tagged B2B or Marketing across every batch, it beats only 24% to 69%.

However, the Spring 2026 number (while the more flattering one) is the less useful one. It measures whether the domain is normal for its age, not whether the company can win business: half that batch records 3 visits a month and an authority score of 3, and the 500 average is a handful of outliers pulling the mean up. Nobody in that room is selling yet.

The B2B/Marketing cohort is the field a buyer actually chooses from. Same problem; products in market, customers to lose. InstaAgent sits below its median. Narrow to the nineteen direct competitors, all with real products and real ages, and it falls into the bottom quarter.

InstaAgent is strong on what it controls. It changed its delivery model when customers said the product alone was not enough, rather than defending the original plan. Its product thesis is specific, and persona-level segmentation is a differentiated answer most of the nineteen-company field does not attempt.

Taking the claims as stated, capital efficiency is the standout figure in this report: $2M on YC's standard $500,000, against rivals reporting $9M and $10M on $23M and $16M raised. That is 4.0x revenue per dollar raised, against 0.63x and 0.39x. If the number holds, it is the best in the set by a wide margin.

What thins is the view from outside. Revenue, client count, and both founders' prior outcomes are self-reported: nowhere contradicted, simply unaccompanied. Branded search registers zero in every month on record, the footprint sits below the industry median on four of five metrics, and we located no pricing page, case study, press mention or open role, while seven of nineteen peers list open roles.

Three paths stay open: a durable relationship-sold agency that never needed a web footprint; a version where the footprint catches up, with branded search, a broader roster and one checkable client arriving together; or one where Meta and TikTok finish shipping the capability and the wedge closes from above. Of marketing companies old enough to have resolved, 43% are still operating, 32% were acquired and 25% went inactive. Nothing in that spread favours one of the three paths over another, which is the point. Everything countable here describes a small, new company doing something specific and doing it early. The one thing this public-source analysis cannot count is what the whole story rests on: the revenue itself. That is less a flaw than the shape of any business sold through relationships rather than a funnel. A bullish future points to a traffic pull with branded search emerging and referring domains crossing the industry median, plus one client relationship becomes independently verifiable (press, a named case study, a client's own social post), a base future shows revenue growing but staying founder-sold and concentrated, web footprint staying thin but stable, pointing to a durable services business rather than a venture-scale platform, while a bearish future leans towards acquisition and/or complete market-policy shift. 

YC has now funded nineteen companies in or adjacent to AI-generated advertising. InstaAgent stands out because its company-reported traction is substantially larger than the public footprint observable through the sources used in this analysis. Public data neither verifies nor disproves those figures, and several ordinary explanations, including a relationship-led sales model and client confidentiality, could account for the difference.

That gap may reflect a young, relationship-led business whose customers and revenue do not surface online. Public sources do not disclose how InstaAgent defines client count, recognizes ARR, or characterizes the relationships represented by the customer logos, so those figures cannot be interpreted more precisely from the available data. Pricing, retention, customer concentration, and one independently verifiable case study would resolve most of the uncertainty. 
Until then, the strongest conclusion is not that the story is wrong; it is that the coming months will show whether the public footprint begins to match the reported traction.