April 16th, 2026

Microsoft Copilot is now inside Power Apps — and the agent feed drops May 4

Tags: #RT-PowerPlatform #ESRA #RT-AI-Role Source: TheTip.ai (Apr 16) | Primary URL: https://techcommunity.microsoft.com/blog/microsoft365copilotblog/copilot-in-word-new-capabilities-for-document-workflows/4508974 Verdict: verified shipped

This isn’t “Copilot beside Power Apps” — it’s embedded inside the app. Same permissions, same business logic, same data context, already live. Ask a question in plain English; the app reshapes the view. Fill a form; AI auto-populates fields from incoming emails. The big news for May 4: an agent feed — a dedicated space where business users see, review, and guide agent actions in real time, with approval thresholds before anything executes.

The headline for Roy’s RT work: Power Apps is now exposing its intelligence outward via MCP servers, so any agent running outside Power Apps — including one you build on Claude — can invoke Power Apps logic and access its records through standard MCP protocol. The documentation tool you built at RT just got a potential live-data channel.

If the RT AI role interview happens, this is a live example of the kind of “enterprise AI embedded in existing workflows” work you’d be doing. You can speak to it concretely: the May 4 agent feed is exactly the oversight layer ESRA still needs.

Action: Mark May 4. Watch the agent feed rollout — this is your adoption-strategy argument made into a Microsoft product. If the RT demo hasn’t happened yet, Power Apps + Copilot is a stronger live reference than what you had last week.

Humwork: AI agents can now hand off to a verified human expert in under 30 seconds

Tags: #MACA #Ben #Always-On-Reeve Source: TLDR AI (Apr 16) | Primary URL: https://www.testingcatalog.com/humwork-a2p-marketplace-connects-ai-agents-with-experts/ Verdict: verified shipped (YC P26)

Humwork launched the first Agent-to-Person (A2P) marketplace. When an AI agent hits something it can’t confidently handle, it can hand off — with full session context — to a verified human expert in under 30 seconds. 1,000+ experts available globally. 87% resolution rate. Already integrates with Claude Code and Replit.

This fills a real gap in MACA’s 14-agent pipeline. Right now, when a wave fails or produces a result you’re unsure about, the fallback is Roy manually reviewing. The A2P pattern is the architectural answer to that: build in a conditional handoff trigger when confidence is below threshold, and a human picks up the thread with full context intact. You don’t have to use Humwork specifically — but the pattern is production-ready and the market just validated it.

Ben has a 3-tier authority model precisely because some decisions should escalate. This is the same principle extended to an external marketplace of domain experts.

Action: Look at the Humwork API docs. Even if you don’t implement it immediately, designing the handoff hook into MACA’s next build sprint costs nothing and makes the demo more credible to a gym owner who asks “what happens when it gets it wrong?”

LinkedIn’s AI hiring agent is growing 36% per week — and its cold messages beat humans by 50%

Tags: #CourseBuilds #Prevail-Partners Source: The Information, Applied AI (Apr 16) | Primary URL: https://www.theinformation.com/newsletters/applied-ai Verdict: verified first-party (LinkedIn CBO Mark Lobosco on record)

LinkedIn launched Hiring Assistant in September 2025. As of today, it’s growing 36% every week — and AI-generated outreach messages are getting ~50% higher response rates than human-written ones. LinkedIn CBO Mark Lobosco compared it to Recruiter, the product that became LinkedIn’s first $1B ARR line. Senior Microsoft executives are now studying it to figure out how to replicate the pattern in other Copilot products.

The stat that matters for Roy: the recruiter liked their own messages better, but the AI’s messages performed better. That’s a real thing you can say to Michael Zaicek at Aria when he asks “can AI actually write better than my team?” Yes. Here’s the data. At scale. On LinkedIn. From the company that sees 200M+ recruiters.

This is also the enterprise adoption proof point you’ve been waiting for. Roy is building CourseBuilds specifically to help businesses like Aria capture this kind of value. The argument is no longer theoretical.

Action: Screenshot the stat. Have this ready for the Zaicek conversation: “36% week-on-week growth, AI messages outperform humans by 50% on reply rate. This is what happened when LinkedIn gave their own team the tools we’d give yours.”

The Information reported today that enterprise CIOs are “bracing for higher Claude bills” after Anthropic’s pricing shift to consumption-based billing. National Life Group’s CIO was quoted saying it “makes it unpredictable within an enterprise model” and they’re sticking with OpenAI because it’s “easier to predict.”

Roy already built a cost tracker into MACA — api/lib/costs.ts, the dashboard at public/cost-dashboard.html, per-run cost logging, per-ad-unit economics. While Fortune 500 CIOs are discovering that consumption-based AI billing is “unpredictable,” you’re already running a 14-agent pipeline with granular cost visibility per wave, per ad unit, per run.

That’s the gap you’re selling into with CourseBuilds and with every MACA demo. Businesses don’t just need AI tools — they need someone who can help them understand what those tools cost at their actual usage scale before rolling out to 53 staff. That’s the wedge Roy has at Aria that a generic SaaS subscription can’t replicate.

Anthropic’s consumption pricing → CIO anxiety is your pitch prep

Tags: #CourseBuilds #MACA Source: The Information Applied AI (Apr 16)

The named quote from National Life Group’s CIO: “There isn’t much history of usage for us to go on to appreciate or understand the change. At some point, pay by the drink gets unaffordable.”

Before the Aria meeting with Zaicek, understand what Claude consumption pricing looks like for a 53-person firm doing moderate AI use. Build a rough estimate. Walk in with “here’s what I’d expect it to cost your team, here’s what the ROI looks like at that spend level.” That kind of pre-work closes deals. CIOs are anxious about this right now — Roy showing up with a prepared cost model is a differentiator.

Cloudflare Browser Run — edge-native AI agent browser infrastructure

Tags: #MACA #Ben Source: TLDR AI (Apr 16) | Primary URL: https://blog.cloudflare.com/browser-run-for-ai-agents/

Cloudflare rebranded Browser Rendering to Browser Run and added purpose-built features for AI agents: isolated execution contexts, structured content extraction, session management built for automation at scale.

Ben already uses Playwright for the recon scraper. MACA’s photo pipeline touches the web. If either project ever needs to scale beyond single-machine execution or you want to move away from a local Playwright dependency, Browser Run is the edge-native alternative. Not urgent today — but worth bookmarking before the next MACA build sprint involves any web scraping.

“Snap just announced they’re cutting 16% of their workforce — 1,000 people — and in the same announcement said AI is now writing 65% of their new code. The stock went up 8% on the news. Wall Street is rewarding AI-driven efficiency, full stop. And the company that sells AI adoption services to businesses like yours is the one helping you get ahead of that curve, not catch up to it.”

Use in: CourseBuilds pitch to Aria, RT AI role interview, anyone asking whether AI is really changing business outcomes yet. The stat is independently verifiable via Snap’s official newsroom.

Source: Snap newsroom + The Information AM (Apr 16) | URL: https://newsroom.snap.com/organizational-changes-at-snap

Allbirds just pivoted to GPU rentals and the stock went up 600%. There’s a UBX angle here.

Tags: #UBX-South-Bank-Sale (framing signal) Source: The Rundown AI + The Information AM (Apr 16) | Primary URL: https://ir.allbirds.com/news-releases/news-release-details/allbirds-inc-executes-50m-convertible-financing-facility Verdict: verified (SEC filing)

Allbirds — the $4B IPO darling that became a $22M micro-cap — announced a $50M convertible note deal to become “NewBird AI,” a GPU rental business. Stock up 600%+ in one day.

The pattern: markets are applying a radical valuation premium to anything with credible AI transformation narrative right now. This is the same thing that happened with crypto rebrands, but with legitimate underlying demand (compute scarcity is real).

Here’s the UBX angle Roy hasn’t considered: the gym sale is currently framed as “operational business, 6-year track record, strong community.” What if the buyer framing included “AI-enabled operations” — MACA ad system, Ben’s autonomous Xero bookkeeping, the PaperClip infrastructure, the data pipeline? You’re not selling a plain gym. You’re selling a small business that already has a working AI operations layer. Right now, that framing carries premium valuation energy.

This doesn’t change the fundamentals of the sale — the business is worth what it earns. But in the buyer narrative and outreach to the corporate UBX pipeline, leading with “the first AI-native UBX franchise” might get better attention than “profitable gym for sale.” Worth one conversation with Roy and Sush before the Week 1 Michael Jordan outreach email ships.

Items covered in prior briefs (appeared in today’s emails, no material update):

• Adobe Firefly AI Assistant — covered Apr 16
• Notion built-in Claude agents guide — covered Apr 16
• OpenAI ChatGPT CPC ads launching — covered Apr 16
• Microsoft Copilot OpenClaw features — covered Apr 16
• Claude Code Routines — covered Apr 16
• Claude Code desktop redesign / parallel sessions — covered Apr 16
• Dharmesh AUX article — covered Apr 16
• Anthropic pricing shift base story — covered Apr 10/16
• Bagel Bots sellable offer prompt — covered Apr 16
• Google Chrome Skills — covered Apr 16

New items considered and skipped:

• OpenAI Agents SDK update — cross-file/tool harness, sandboxed execution. Anthropic is Roy’s stack; no project hook beyond staying aware. Skip.
• Gemini Mac desktop app — shipped, late to market vs Claude. Roy uses Claude desktop natively. Conversation context only, no action needed.
• Starbucks ChatGPT integration — good “AI inside everyday apps” pattern, but no project hook. Conversation capital at best, but Snap stat already fills that slot.
• GPT-5.4 Pro solves 60-year math problem — impressive, zero project relevance.
• Gemini 3.1 Flash TTS — not in Roy’s audio/TTS space. Skip.
• TSMC +30% revenue forecast — macro AI chip demand signal. No actionable step this week.
• Salesforce Headless 360 / MCP — important enterprise MCP adoption signal, but Salesforce isn’t in Roy’s client stack. Direction-of-travel bar not cleared.
• Anthropic design tool / Mike Krieger leaves Figma board — material update to the Opus 4.7 / design tool story (covered Apr 16). The design tool could eventually help with UBX sale site and Prevail website builds. Watch for when it ships. Not actionable today.
• Neil Patel — “AI is changing paid search” — CPCs rising, CTR dropping. Generic framing with no specific tool or strategy. Skip.
• a16z — “Prediction Markets: They Grow Up So Fast” — not AI. Skip.
• Claude “probably wasn’t secretly nerfed” — useful diagnostic explainer (cache duration, context compaction, quota policy = experience changes without model changes). Low urgency since Roy already has cost tracking in MACA. Bumped to skip.

• Newsletters scanned: 9 sources, 13 individual emails read
• Items extracted: 17 distinct new AI-related claims/tools/announcements
• Already in covered-stories: 10 items
• New items evaluated: 17
• Items surfaced: 3 (Section 1) + 2 (Section 3) + 1 conversation capital + 1 first-mover = 7 total
• Items newly skipped: 14
• Read time: ~5 minutes