Daily Briefing
June 24, 2026 · 5 items (site) · 6 items (base)
🔥 Headlines
01
Zafin launches AIOS — banks finally get a switchboard for their AI agents
Bank software vendor Zafin ships AIOS, the first platform designed to coordinate a bank's own AI agents, partner agents, and any outside agent a regulator might allow — while keeping a clean record of who did what, with which model, at what cost, and where a human stepped in. Think of it as the air-traffic-control tower for an industry that, until now, had no way to supervise multiple autonomous agents running side by side. The timing is no accident: Deloitte says only 21% of organizations have a mature governance model for autonomous agents. AIOS is built to fill that gap. The platform is already live inside Zafin itself, where it runs the company's own software delivery and modernization work before being sold to clients.
02
Engram raises $98M to make enterprise AI agents 100× cheaper to run
Engram, an 8-month-old, 13-person startup, raises $98M from General Catalyst, Kleiner Perkins, Sequoia, and Andrej Karpathy himself. The product sounds dull — "memory models for organizations" — but the result is dramatic: clients like Microsoft, Notion, and Harvey get the same answer quality as a frontier model for roughly 1% of the token cost. The trick: a learned memory layer per organization means the model doesn't have to relearn your company from scratch on every prompt. This is the clearest sign yet that the agent boom has entered a sober phase — when the bill for inference is bigger than the productivity gain, the buyer pauses. Engram's pitch is the pause-buster.
03
Journey.ai × Dialpad — AI agents can finally take a card payment without seeing the card
At Customer Contact Week in Las Vegas, Journey.ai and Dialpad announced a partnership that solves one of the messiest problems in agentic commerce: how do you let a conversational AI accept a payment by card without the card number ever passing through the AI, the call recording, or the agent's environment? Journey's Zero Knowledge® layer opens a secure window next to the chat; the customer types their details there, the agent simply sees a green checkmark. The two systems talk via MCP, the emerging standard for agents calling tools. The broader pattern: the AI acts, a human supervises, sensitive data never leaks into the AI pipeline. That pattern is becoming a category.
04
Kanverse.ai ships a no-code platform that turns finance teams into agent builders
San Jose-based Kanverse.ai is moving from "smart document reader" to end-to-end agentic finance platform. The new piece: an AI Agent Studio where a finance operator — not a developer — describes in plain language what they want automated (read this PDF, check this rule, escalate if over $10k, post to the ERP), and Kanverse ships a working agent that does it. Target: corporate finance departments that handle a flood of invoices, purchase orders, and approval chains — and don't have an engineering team. CEO Karan Yaramada: "Companies are past the AI experiment phase. They want AI to do real work, safely, across processes." The no-code claim is the moat against UiPath, Workday, and a wave of agentic-finance startups.
05
e& UAE + TM Forum publish a real roadmap for self-running telecom networks by 2030
At DTW Ignite in Copenhagen, e& UAE (the UAE's biggest telco) and the industry body TM Forum released a white paper that does something rare: put a date and a level on the march toward fully agent-operated telecom networks. Target: autonomy level 4 by 2030 — a network that detects, diagnoses, and fixes its own incidents, with humans only on the exceptions. Five building blocks: a multi-domain view of the whole network (RAN, core, transport, IP, fixed), AI-native operations that work on any vendor's gear, open-source agent tooling, end-to-end closed loops, and a customer-experience layer that reasons about subscribers, not just boxes. Critical infrastructure is going agentic — on a published schedule, not as a vague ambition.
📡 To Watch
Agent governance is becoming a product category of its own
Zafin AIOS (banks), Microsoft Agent Framework (BUILD 2026), and the June 23 cohort of Ent / NewCore / Arcade ($226M combined) all attack the same hole: who watches the agents? Expect a wave of "agent control plane" startups, plus the first NIST and OWASP guidelines specifically for agent identity and authorization.
Inference cost is the new bottleneck — memory layers are a market
Engram's $98M is the loudest signal yet that the agent boom is splitting into two speeds: frontier-model performance for the demos, and specialized memory layers for production. Watch for copycat raises in "learned-memory-as-a-service" through H2 2026.
How does an agent pay another agent? Two startups, two answers
Journey × Dialpad solves the human-facing side (PCI payments during a call). Orthogonal (Pantera, YC, $4.3M seed) solves the agent-to-agent side (multi-chain, no human, no bank). Both are early, both are tiny — but the underlying question ("what's the financial rail of the agent economy?") is now on every cloud vendor's whiteboard.
Microsoft Copilot Cowork GA — the "model-picker" pattern goes mainstream
Microsoft rolls out Copilot Cowork to all M365 tenants, with a model selector (Auto / Claude Opus / Sonnet / GPT-5.5) behind a single agent surface. If it sticks, it normalizes the idea that one agent can swap the brain inside it — and that model choice is a user-facing feature, not a developer secret.
📊 Trend
June 24, 2026 looks like the day the agent stack went vertical. Five sectors moved at once: banking (Zafin AIOS ships a control plane for regulated agents), cost (Engram's $98M makes specialized memory a category), commerce (Journey × Dialpad prove an agent can take a payment without ever seeing the card), finance operations (Kanverse turns AP clerks into agent builders without code), and critical infrastructure (e& UAE + TM Forum commit to autonomy level 4 for telecom by 2030). The common thread: every layer of the agent stack — coordination, memory, payments, build tools, deployment — is getting its own dedicated product, its own dedicated funding, and its own dedicated category of buyer. The agent economy isn't a single market anymore; it's a stack of tightly named sub-markets, each with a leader emerging in real time.