The archive · AI & Models · Product decision · 2026
Tobira bets AI agents need @addresses: PH #1 launch, then 593 agents and a leaky funnel
Porto's Vlad Shipilov bets agents need their own networking layer — public @addresses and agent-to-agent dealmaking; PH #1 launch, 593 agents by April 2026.
Tobira
What the business is
Tobira is an open network of public addresses for AI agents — like email but for AI — where agents discover each other, negotiate, and find clients, partners and deals on behalf of their human owners.
How it started
Vlad Shipilov, a serial entrepreneur based in Porto, founded Tobira after concluding agents could do everything except find the right people. He launched on Product Hunt on March 23, 2026 as a free open network compatible with OpenClaw and Claude Cowork, two weeks after Meta acquired Moltbook, the viral 'Reddit for AI bots'.
What happened
The launch worked: #1 Product of the Day with ~596 upvotes, 470+ agents and 4,200+ conversations within five days. Shipilov then published unusually candid first-party analytics: of 4,882 conversations only 6.7% reached fact-check, 0.2% reached deep dialogue, and 87% of escalations to human owners expired unanswered — evidence that matching volume was not converting into human handshakes.
How it ended up
Still live and building: as of May 2026 Tobira is free, with no disclosed funding or revenue, and the founder is publicly shipping fixes for onboarding drop-off and owner re-engagement.
Background
Tobira was launched on March 23, 2026 by Vlad Shipilov, a serial entrepreneur in Porto, Portugal, on the thesis that the agent economy lacks an addressing and discovery layer. Most agent platforms focus on what agents can do; Tobira focuses on who they can talk to, giving each agent a public @handle, a structured profile of offers and needs, and the ability to negotiate with other agents before any human contact details are exchanged.
The launch landed: Tobira took #1 Product of the Day on Product Hunt with roughly 596 upvotes, and within five days reported 470+ live agents, 4,200+ real conversations and agents from 20+ countries — founders, investors, recruiters and consultants. The positioning was deliberately commercial rather than social: every interaction carries business intent, closer to LinkedIn than to Moltbook, the viral 'Reddit for AI bots' Meta had just acquired.
Shipilov then published first-party funnel analytics that showed the gap between volume and outcomes: of 4,882 conversations started by April 6, only 6.7% reached fact-check and 0.2% reached deep dialogue, while 87% of escalations to human owners expired without a response. He attributed the collapse to onboarding drop-off (62%), owners pausing agents, and trust cold-start — framing the matching algorithm as solved and human re-engagement as the open problem.
What has to be true
- The agent economy was growing fast but agents had no way to discover each other — Tobira attacked the addressing layer, the piece Shipilov argues MCP and A2A forgot.
- Launching two weeks after Meta acquired Moltbook for its 2.8M-registered-agent network gave the 'agents need a network' thesis mainstream validation.
- A privacy-first consent model — no contact exchange until both sides approve — differentiated Tobira from marketplaces where humans browse and hire agents.
- Publishing raw funnel data was a trust play: showing 4,882 conversations but only 0.2% deep dialogue built credibility with builders even as it exposed the product's immaturity.
- Free, open, framework-compatible onboarding (OpenClaw, Claude Cowork) lowered the cost of joining a chicken-and-egg network.
What can be applied
Network launches produce volume, not trust: Tobira got 4,882 agent conversations but almost no human handshakes — for agent networks the bottleneck is human re-engagement, not matching.
Aftermath
As of May 2026 Tobira is live and free, with no disclosed funding or revenue and a small indie-builder operation. The founder's public roadmap from the April 6 analytics targets the funnel's bottlenecks: rewriting the onboarding agent to cut the 62% drop-off, nudging owners to resume paused conversations, and pre-loading credibility for new agents. The platform remains compatible with OpenClaw and Claude Cowork, and the category — agent-to-agent discovery — is heating up as Google, Mastercard and Microsoft push agentic commerce protocols.
Sources
- AI matching conversion rate: what 4,256 matches and 4,882 conversations taught us about funnel design
- What If Your AI Agent Had Its Own @Address? Tobira.ai Is Building That Network
- Tobira — 1 AI Tool
spotted an error? The archive wants to know.
Your turn
You just read one. Describe what you are building, and see who is betting on the same thing.
Free account · 3 free questions · no card