Tech Digest — May 22, 2026

Top story

Google ships Gemini Spark — a 24/7 agent that keeps working while your laptop is closed

Why it matters for entrepreneurs: Spark is the first always-on agent from a major platform built to own a workflow end to end, not answer a question. It runs on dedicated cloud machines, so it keeps working when your devices are off, and it will connect to outside tools through MCP (Model Context Protocol, the open standard that lets agents talk to apps) in the coming weeks. This is a preview of what your clients will soon expect "AI" to mean — and a benchmark your own service has to clear.

Spark is in trusted-tester mode now, with a beta for Google AI Ultra subscribers in the US next week. The roadmap matters more than the launch: custom sub-agents, email and text control, and authorized payments with budget and merchant limits, all promised through the summer. Treat that as a deadline, not a demo. The same MCP plumbing that lets Spark act is the plumbing you can wire into client automations today — the gap between "Google's agent" and "your agency's agent" is mostly integration work, not model access.

Quick hits

Notion turns its workspace into a hub for AI agents

Source TechCrunch

Notion's new Developer Platform adds Workers (a sandbox for running custom code), an External Agents API that runs Claude, Codex, and Decagon agents out of the box, and Database Sync that pulls live data from Salesforce, Zendesk, or Postgres into a Notion database. It is a bid to be the orchestration layer where people and agents share one workspace — and Workers are free through August, so testing it costs nothing but time.

Zendesk pushes "billed per resolution, not per seat"

Source CMSWire

At its Relate 2026 conference, Zendesk leaned hard into outcome-based pricing: AI agents that cost roughly $1.50 to $2.00 per verified resolution instead of a per-seat license. The per-resolution model is not new — it shipped in 2024 — but the "autonomous AI workforce" framing is, and it is a working example of AI shifting the billable unit from access to outcomes.

Anthropic keeps its cyber-capable model on a short leash

Source Just Security · Computing

Mythos — a model strong enough at finding and exploiting software vulnerabilities that Anthropic will not release it publicly — stays limited to roughly 50 vetted partners under "Project Glasswing," though Anthropic softened some vulnerability-reporting restrictions this month. The signal for smaller firms: offensive AI capability is arriving faster than the defenses, and patching fast is becoming a survival skill rather than an IT chore.

Tool / launch watch

Gemini 3.5 Flash. Alongside Spark, Google previewed Gemini 3.5 Flash, a lighter model it says runs at roughly half — in some cases close to a third — the price of comparable frontier models. For an agency, the cost line is the story. Cheaper per-token inference makes always-on automations affordable to leave running: inbox triage, lead scoring, draft generation, monitoring. It is worth re-pricing any workflow you shelved because the model bill did not pencil out.

Funding / M&A pulse

Hark raised over $700M at a $6B valuation. The AI hardware startup's Series A was led by Parkway Venture Capital, with Nvidia, AMD, Intel, Qualcomm, and Salesforce all on the cap table. When chip rivals co-invest in the same hardware startup, the bet is on overall supply, not on picking one winner.
Martha Stewart's Hint raised a $10M seed from Slow Ventures. The startup points an AI assistant at home maintenance — insurance, utilities, repairs. Vertical AI for un-glamorous, recurring household work is exactly the wedge SMBs in trades and home services should be watching.

Angle for the blog

Headline: "Bill the Outcome, Not the Hour: What Zendesk's Pricing Move Means for AI Consultants." This week's news has one practical lesson for anyone selling AI services, and it is not about Gemini — it is Zendesk charging per resolved ticket. As agents shift from answering to acting, the honest billable unit shifts with them: away from hours and seats, toward outcomes a client can count, like appointments booked, leads qualified, or tickets closed. A 600-word post would walk a service firm through repricing one offer around an outcome — pick a workflow with a countable result, instrument it so the result is verifiable, set a per-unit price that beats the client's current cost, and keep a floor retainer to cover the build. The contrarian edge: outcome pricing is usually sold as a win for the buyer, but it is a bigger win for the seller who actually trusts their automation — it uncaps revenue on your best work and quietly fires you from your worst. Tie it to the "Simplify It" promise: clients do not buy agents, they buy resolved work. Keep this a seed — leave the build steps for the post.

The Tech Digest is compiled each morning by SyncBroad AI — a plain-English read on AI for service businesses. Browse the full archive, or book a 15-minute demo to see what's actually deployable for your operation.