Tech Digest hero — May 27, 2026

Top story

Anthropic's MCP tunnels let agents reach inside your network without opening a firewall

Source Anthropic

Why it matters for entrepreneurs: Anthropic added two features to Claude Managed Agents on May 19: self-hosted sandboxes (public beta) and MCP tunnels (research preview). Self-hosted sandboxes move tool execution off Anthropic's infrastructure and onto the customer's own VPC — or onto a managed runtime like Cloudflare, Daytona, Modal, or Vercel. MCP tunnels let an agent reach an internal database, ticketing system, or private API through a single outbound connection from a lightweight gateway, with no inbound firewall rules and no public endpoint. The orchestration loop still runs on Anthropic, but the data and the code that touches the data stay where compliance already approved them.

The practical read for any agency selling AI into regulated buyers — healthcare, legal, defense, financial services, government contractors — is that the strongest version of the "data can't leave our perimeter" objection just lost its teeth. You can now sell a managed agent to a client whose security team has spent five years rejecting outbound SaaS exceptions, because nothing leaves the perimeter except the agent's planning calls. Pair it with the existing audit logging and you have a story the CISO can actually sign off on. Expect the bigger platform vendors (OpenAI, Google Vertex) to ship parity features within 60-90 days. The window for being early on this is narrow.

Quick hits

Salesforce drops Agentforce Coworker into every search bar

Source Salesforce Ben

Announced by Benioff on May 21, Coworker turns Salesforce Global Search into a conversational agent that can summarize records, surface insights, and take action — and it also runs natively inside Slack, Teams, and ChatGPT. For agencies serving Salesforce-anchored clients, the takeaway is that custom Agentforce builds just got more useful for sales reps who never opened the actual Salesforce UI, which removes one of the bigger ROI objections to Agentforce projects.

Camunda ships ProcessOS — process discovery as a service

Source Camunda

ProcessOS, in closed beta May 20, watches how work actually moves through a business, then converts described outcomes into governed agentic workflows with human review built in. For consultancies that bill discovery hours to map a client's broken process before rebuilding it in Make.com or n8n, this is either a threat or a leverage tool, depending on whether you front-run the change. Worth a closed-beta application now.

NVIDIA-verified agent skills bring code-signing to the agent layer

Source NVIDIA

NVIDIA published a pipeline on May 19 that catalogs, scans, signs, and documents portable agent skill packages. Procurement, security, and SRE teams can now assess a skill against a manifest before approving deployment — the same pattern Linux distros and container registries have used for years. As soon as one big enterprise standardizes on "we only deploy signed skills," every reseller and agency in their orbit has to follow. Build with portability in mind.

Anthropic in talks for $30B at $900B valuation

Source Bloomberg

Three months after closing the $30B Series G at $380B, Anthropic is reportedly raising another $30B at a valuation north of $900B. The signal for the market: the lab-tier race is now firmly a two-horse compute story (Anthropic + OpenAI), which means pricing on Claude and ChatGPT API tiers is unlikely to drop the way mid-market vendors keep promising. Lock in your unit economics on the assumption that prices stay flat or drift up, not down.

Tool / launch watch

DocuSign Iris AI agents (announced May 21 at Momentum) handle the full agreement lifecycle — drafting, redline review, routing, post-signature obligation tracking. For consultancies and agencies that send a lot of SOWs and renewal MSAs, this is the first DocuSign-native option that's competitive with the standalone CLM startups. Worth a 60-minute hands-on if your contract ops are still half-manual. Pricing tiers haven't been published yet, so don't quote a client until they are.

Funding / M&A pulse

Angle for the blog

Headline: "The 'our data can't leave the building' objection just died — here's the playbook." The angle is for any agency or consultancy that's lost a deal to a client's security team in the past year. Open with the specific objection — "managed AI means our customer data sits on someone else's servers, and our compliance officer will not sign off." Walk through what Anthropic's MCP tunnels and self-hosted sandboxes actually change: tool execution moves to the client's VPC, internal databases become agent-callable without a single inbound firewall rule, and the only thing crossing the perimeter is the agent's planning loop. Translate this into three concrete consulting plays for the next 90 days — (1) re-open the deals that died on the data-perimeter objection with a one-page architecture diagram showing what now stays inside; (2) productize a "Compliant Agent Starter" that bundles a self-hosted sandbox in the client's existing AWS or Azure account with a vetted set of MCP tunnels into their CRM, ticketing, and document systems; (3) write the security one-pager the client's CISO can hand to their auditor without redlines. Close with SyncBroad AI's positioning: we don't ask your clients to make a SaaS exception — we build the agent inside the perimeter they've already approved. "Simplify It" means making the secure path the easy path, not the slow one.

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.