EarlySignals Weekly

Tuesday, May 26, 2026

Every firm deep-researched by agentic AI, scored for potential and scoop, and fact-checked before publish.

40 firms: AI agents acquire dedicated infrastructure, approvals, and vertical deployments.

Agent payments and messaging gain scoped controls. Allowance now routes every AI agent transaction through mandatory iPhone approval after its solo founder faced raw card exposure, while Chert sells the programmable iMessage infrastructure that first reached 2,000 users in a week.

Voice agents attach to existing PBX systems. Callab AI ships agents that connect to Avaya and CUCM via native SIP trunks, already serving Dunkin Donuts across three continents.

Insurance carriers underwrite agent deployments. Mount automates underwriting and claims for AI agents and offers an Agent Deployment Readiness certification, led by two Swiss founders who previously published on object detection.

Vertical agents cut operating costs in media, rentals, and security. Aspect builds agents for media asset pipelines, ProhostAI automates 70 percent of guest messages for short-term rentals, and Ocean runs specialized agent swarms that already protect KAYAK and Headspace inboxes.

The remaining firms include quantum error-correction hardware, cultured dairy, open European language models, and Middle East proptech tools.

ShareXLinkedIn

You are reading 5 of 40 firms in this edition. Get all 40 for $29/month, free for 14 days.

Unlock the full edition

5 of 5

Scores run 0 to 100. Potential: how likely the firm is to become a real winner, judged across the six dimension scores and held consistent with them, whether or not anyone has heard of it. Scoop: how unlikely a well-read tech investor is to have heard of it yet. High means it has not reached the outlets that reader follows, whatever the firm's quality or age. How our top potential scores did.

Raised (1)

Ocean

Verified

potential 82 · scoop 28

Tel Aviv, Israel (with New York office)Founded 2024 · Announced May 19, 2026 (seed funding)Surfaced May 26, 2026

Ocean runs a swarm of specialized agents that investigate incoming email for intent and context. The company protects hundreds of thousands of mailboxes at KAYAK, Headspace, and Kingston Technology after raising $28 million.

Shay Shwartz and Oran Moyal, who met at sixteen in Israel's Magshimim program, raised $28M for an agentic platform already shielding KAYAK and Headspace.
Team 85 Product 75 Traction 65 Market 90 Moat 70 Capital 85

Launched (2)

Aspect

Verified

potential 42 · scoop 82

USSurfaced May 26, 2026

Aspect builds AI agents for media asset management and desktop streaming into editing tools. The YC F25 startup comes from three founders who previously scaled pipelines at rival Shade.

Three University of Michigan CS graduates who formed Shade's early team now compete against their former employer after launching Aspect in the YC F25 batch.
Team 65 Product 70 Traction 25 Market 55 Moat 40 Capital 45

Hessian

Verified

potential 68 · scoop 85

USSurfaced May 26, 2026

Hessian deploys custom AI agents inside client operations. Three Durham CS graduates with Bloomberg and GoCardless experience target the gap between pilot tools and production workflows.

The team behind a zero-downtime migration of a payments system processing over one billion pounds on peak days now embeds to build reliable AI agents for enterprises.
Team 80 Product 70 Traction 35 Market 75 Moat 55 Capital 65

Building (2)

Bielik (SpeakLeash Foundation)

Verified

potential 48 · scoop 85

Warsaw, PolandAnnounced May 1, 2026 (pre company build)Surfaced May 26, 2026

SpeakLeash Foundation released Bielik, an 11B Polish LLM trained on about 200B tokens over two epochs. The model leads Polish benchmarks through a custom APT4 tokenizer and national HPC resources.

Polish volunteers trained Bielik 11B on Cyfronet AGH supercomputers and posted top scores on Polish benchmarks within months of the 7B release.
Team 68 Product 82 Traction 74 Market 87 Moat 62 Capital 35

UK-LLM (UCL Sovereign AI Initiative)

Verified

potential 58 · scoop 78

London, United KingdomAnnounced Mar 1, 2026 (pre company build)Surfaced May 26, 2026

UCL NLP researchers post-trained NVIDIA Nemotron models on Isambard-AI using 30M+ synthetic Welsh translations. Their earlier Caernarfon 3B model tops BritEval Irish but trails Llama 3 8B and Mistral 7B on Welsh.

Pontus Stenetorp's UCL group post-trained Nemotron bases on Isambard-AI for Welsh, and its earlier Caernarfon 3B beat larger models on BritEval Irish.
Team 80 Product 70 Traction 45 Market 75 Moat 55 Capital 65

35 more firms surfaced this week

10 from United States, 9 from Canada, 9 from Israel & Middle East and 7 from Europe.

Subscribe to read the full edition: every firm's deep research, potential and scoop scores, the full dimension radar, all fact-checked before publish.

Free for 14 days, then $29/month, or $290/year (2 months free). Cancel anytime.