The Agent-to-Agent Pitch Is Coming for Your DMs Next
Every social media operator I know is quietly running the same experiment right now: pointing an AI agent at an inbox and seeing whether it can do the boring half of the job — the triage, the first reply, the “is this even worth my time” filter. So when a founder launches an agent-to-agent network that pitches 2,000 VC funds on your behalf and returns YES, NO, or MAYBE, I don’t read it as a fundraising story. I read it as a preview of what’s about to happen to creator brand deals, podcast guesting, and cold outreach across every platform we run. Bleetz Network, built by Davit Svanidze and hosted on Vercel, is a small, free, slightly unhinged proof of concept — and it’s worth studying precisely because it’s not polished.
What Bleetz actually does — and why the framing matters more than the product
The pitch, per the maker’s own launch post: Bleetz is “the world’s first (and, if I’m not mistaken, only) A2A (agent-to-agent) network,” where businesses find suppliers and customers in seconds. For this specific launch, the vertical is fundraising. The team scraped and enriched 2,000 VC funds, turned each into a “simulated VC agent,” and lets your startup agent pitch them. You get one of three answers — YES, NO, or MAYBE — and a YES unlocks the fund’s contact details and investment team. The platform is completely free, which the maker frames as FlowMarket’s gift to founders.
Here’s the part most coverage will skip: the maker is honest about the seams. When asked how the ten-round agent conversation works in the messy real world, Svanidze replied that they “in fact reduced them to 1-2 rounds, since more rounds caused more hallucination and reasons for false negatives.” That’s a real engineering admission, not marketing copy. He also noted the VC agents are “bit too ‘kind’ right now, the ‘YES’ share is too large” — and that they haven’t spoken with many VCs because the platform went live the day before.
My take: that candor is the most valuable thing on the page. It tells you exactly where the ceiling is.
Why TikTok creators should care more than LinkedIn ones
Fundraising feels distant if you’re a solo creator. But swap the nouns. Replace “2,000 VC funds” with “2,000 brand partnership managers.” Replace “startup agent” with “creator media kit agent.” Replace “YES/NO/MAYBE” with “send rate card / pass / ask for audience demo.” The mechanic is identical, and the platforms where this lands first are the ones where cold outreach already dominates: LinkedIn for B2B and creator-economy partnerships, and email for everything else. TikTok and Instagram creators get pitched constantly by DTC brands; the reverse — creators pitching brands at scale — is still done by hand, badly, in spreadsheets. That’s the gap.
The real problem: outreach is a matching problem wearing a volume costume
Most creator tooling solves the wrong half of this. Buffer, Hootsuite, Later, and Metricool are excellent at the distribution layer — scheduling, queueing, cross-posting, basic analytics. They are not matchmakers. When I scheduled 30 posts across five platforms last month, the tools handled the calendar flawlessly and offered zero help on the harder question: which of the 400 brands in my outreach sheet actually fit my audience, and which are a waste of a follow-up?
Bleetz’s answer is a deterministic filter plus thesis comparison. As the maker explained, “the deterministic filter and thesis comparison already makes sure the targeting is pretty accurate,” and each fund profile shows geography, thesis, stage, and sector. That’s the right architecture. You filter on hard constraints first — sector, stage, geography — then let the LLM handle the fuzzy conversational layer. If you invert that order and let the model do the matching, you get the “spray and pray” behavior the maker explicitly criticized. He called out tools that “spray and pray, basically turning a pitch into a spam email,” and positioned Bleetz against them.
For social media operators, the lesson is: your outreach stack should look like a funnel with a deterministic gate, not a chatbot with a send button.
Where the math breaks
Free is not a business model; it’s a subsidy with a countdown. The maker was transparent: Bleetz stays free “as long as token burn allows it,” and the answer to “does it cost something” was “it doesn’t cost anything, at least, as long as token burn allows it.” That’s a real constraint. Agent-to-agent conversations burn inference tokens on every round, for every user, with no revenue attached. The moment volume scales, either pricing appears, rate limits appear, or the “YES” rate gets tuned down to reduce follow-up rounds — and that last one changes the product’s feel entirely.
There’s also a matching-quality ceiling nobody has solved. The maker admitted hallucination isn’t gone: “It’s not 100% hallucination free, but it fell drastically. Mostly it’s VCs, who are inventing wrong reasons for ‘no.’” A false negative in fundraising is invisible and expensive — you never learn the fund would have said yes. In creator outreach, the same failure mode applies: your agent pitches a brand, the brand’s agent (or the human reading it) invents a reason to pass, and you never see the deal you lost. That’s a trust problem, not a technology problem, and it’s the thing I’d watch hardest.
What creators and social teams should actually borrow from this
Three transferable ideas, ranked by how fast I’d implement them.
1. Simulate the counterparty before you pitch. The most interesting design choice in Bleetz isn’t the matching — it’s that the VC side is a simulated agent you can rehearse against. You pitch, it responds, you see the questions it asks. For creators, the equivalent is a brand-partnership simulator: feed it your media kit, your last 20 posts, your audience demographics, and a specific brand’s public positioning, then have it interrogate your pitch the way a skeptical marketing manager would. You’ll find the weak slide in your deck before the real human does. Tools like Canva and CapCut have made the output half of content nearly free; the preparation half is still manual, and this is where the next wave of leverage sits.
2. Make the “MAYBE” a first-class state. The maker defended the three-way outcome well: “Maybe is just a case for the fund agent has more questions, otherwise yes or no are relatively accurate.” Most outreach workflows are binary — sent, or not sent; replied, or not replied. A MAYBE state forces you to build a follow-up question bank. For social teams running influencer seeding, that’s the difference between a flat yes/no pipeline and one where you actually learn why a creator declined. I’d bet the MAYBE bucket is where 60%+ of real conversion lives, and almost nobody instruments it.
3. Let the human claim the agent. Svanidze noted funds “can claim the agents” — the simulated profile becomes a real inbox the fund controls. That’s a clever trust bridge: the agent is a placeholder until the human opts in. For creator outreach, imagine pitching a brand and having the brand’s social lead claim the conversation thread when they’re ready. It reframes cold outreach as an opt-in handoff rather than an interruption. In my experience, the single biggest drop-off in creator-brand deals is the handoff from “brand likes the content” to “brand’s legal/finance person is now involved.” Anything that formalizes that handoff is worth testing.
Where I think Bleetz falls short — and who shouldn’t bother
Let me be direct, because the launch page is candid enough to deserve a candid read.
The VC side is scraped, not consented. The maker confirmed: “The VCs are scraped and enriched, they are simulated, but funds can claim the agents. We notify funds that they have pitches in the inbox.” So you’re pitching a simulation of a fund that may never see your pitch unless someone claims the agent. That’s a meaningful asterisk on the “2,000 VC funds” headline. It’s not fraud — it’s disclosed — but a founder reading “2,000 funds” and imagining 2,000 humans reading their deck is going to be disappointed.
The feedback loop is thin. Asked whether there’s post-pitch feedback collection, the maker described short yes/no feedback and follow-up questions for MAYBE. There’s no mention of structured analytics, no “your pitch scored X on clarity,” no benchmarking against similar startups. For a tool whose whole value prop is speed, the absence of a learning loop is the biggest gap. Compare that to what Notion or Airtable give you for a manual pipeline — at least you can see your own conversion rates.
It is explicitly not for you if: you’re raising a round where warm intros and relationship-building matter more than volume (most seed and pre-seed rounds, honestly); you’re a creator with fewer than, say, a few thousand engaged followers, because brand agents will filter you out on audience size before the conversation starts; or you need auditable, consent-based outreach for compliance reasons. And if you’re a social media manager at a brand, this is a “watch” not a “buy” — the underlying pattern is what matters, not this specific implementation.
The uncomfortable question the launch page raises
If agents pitch agents, and agents decide, where does the human relationship go? The maker’s own answer — “we want our users to have fun” — is charming but evasive. The honest version is that A2A networks optimize for volume and speed, and relationships optimize for trust and depth. Both matter. The winners in the next 18 months will be the tools that use agents to clear the noise so humans can spend their limited attention on the 5% of conversations that deserve it. Bleetz is a crude first draft of that. Crude first drafts are how you learn.
What I’d watch / test next
This week, if you run any kind of outreach — brand deals, podcast guesting, collab requests, B2B partnerships — do three things. First, audit your last 50 outbound messages and tag each as YES, NO, or MAYBE-equivalent. I’d bet you’ll find you’ve been treating MAYBEs as NOs and leaving money on the table. Second, build a one-page “counterparty simulator” prompt: paste in your target’s public positioning and your own pitch, and ask an LLM to generate the five hardest questions they’d ask. Do this for your next ten targets before you send anything. Third, watch whether Bleetz adds a consent or claim-rate metric to its public stats — that number, not the “2,000 funds” headline, will tell you whether the A2A model has legs.
For the broader creator stack: keep your scheduling layer boring and reliable (Buffer, Metricool — pick one, stop switching), and start treating your outreach and matching layer as the place where AI actually earns its keep. The distribution problem is solved. The matching problem is wide open, and Bleetz — free, hallucination-prone, honest about both — is a useful map of the terrain.
