• Anthropic is on track to close a $30B round at a $900B valuation as soon as this week, with Q2 revenue projected at $10.9B (up 130% from $4.8B in Q1) and an expected $559M operating profit — its first profitable quarter, well ahead of prior 2028 guidance. The company also shipped Claude inside Microsoft Word via M365 Copilot for Premium users and used its Code with Claude London event to launch self-hosted sandboxes (public beta) and MCP tunnels (research preview), both letting enterprises run Claude agents inside their own VPC. OpenAI co-founder Andrej Karpathy joined the pretraining team under Nick Joseph to lead work on using Claude to accelerate Anthropic's own pretraining research. (Bloomberg, NYT, Anthropic, InfoQ, TechCrunch, CNBC)

• President Trump postponed signing a pending AI cybersecurity executive order, citing concerns that certain provisions could slow US progress against China. No new signing date has been set, leaving federal AI security guidance in a holding pattern for now. (CNBC, Axios)

• EY and Microsoft committed $1B over five years to embed Microsoft engineers alongside EY consultants and move Fortune 500 AI pilots into governed production across finance, tax, HR, risk, and supply chain. EY will extend M365 Copilot from 150,000 users to its full 400,000-employee workforce. The deal puts EY-plus-Microsoft squarely against Accenture and Deloitte for AI implementation work. (Microsoft News)

• JPMorgan reclassified AI spending as core infrastructure inside its $19.8B 2026 tech budget, alongside data centers and payment systems, with a roughly $2B annual AI line item. CEO Jamie Dimon says the investments have already self-funded through about $2B in operational savings, 500+ live use cases, and a 95% drop in anti-money-laundering false positives. (Fortune, Yahoo Finance)

• Nvidia reported record Q1 FY27 revenue of $81.6B with data center revenue of $75.2B, up 92% year-over-year, on continued AI chip demand. The company said its next-generation Vera Rubin AI chip remains on track for the second half of 2026, starting in Q3. (Nvidia, The Verge)

• OpenAI launched a ChatGPT integration for Microsoft PowerPoint in beta, letting users create and edit decks via chatbot prompts alongside their own documents and images. Available across Business, Enterprise, Edu, Pro, Plus, and Free plans, it extends the Excel and Google Sheets integrations OpenAI shipped earlier this year. (OpenAI, The Verge)

• DeepMind CEO Demis Hassabis nudged his AGI timeline forward to 2029, but the more useful signal for business leaders is what he called the next year: a practical "agentic era" where AI agents handle multi-step workflows inside companies, not a distant breakthrough. Treat the next twelve months as the window to get agent-ready. (Axios)

• Macquarie Bank says it saved 130,000 hours in seven months of Gemini Enterprise rollout, with roughly 80% of its 5,000 staff using it daily. Non-technical teams in legal, marketing, compliance, and sales built "hundreds of solutions" in an internal hackathon weeks after launch. Caveat: figures come from a Macquarie executive speaking at Google Cloud Next, not an independent audit. (iTnews)

• Gartner's 1Q26 survey of 12,004 employees and managers across 40 countries found employees proficient with AI across multiple use cases are 2x more likely to be highly productive, 2.3x more likely to produce high-quality work, and 3.2x more likely to drive process improvements. A positive outlook toward AI adds a 3.4x productivity multiplier on top. (Gartner, May 13, 2026)

• PwC's 2026 Digital Trends in Operations survey of 767 US operations and supply chain leaders found 83% say AI agents and automation will accelerate the breakdown of traditional functional silos. Only 27% have fully embedded an AI strategy across business units, and 37% are comfortable assigning AI agents to full end-to-end processes. (PwC, April 23, 2026)

• BCG's microeconomic modeling of the US labor force found AI is more likely to reshape work than eliminate it: 50-55% of US jobs will be significantly reshaped over the next 2-3 years, while only 10-15% face full displacement over five years. Most reshaping shifts work toward higher-value tasks within existing roles. (BCG Henderson Institute, April 3, 2026)

Proficient AI use compounds; shallow use stalls. This prompt surfaces the repeating tasks in your week and routes each to its best action: save it, try it, or leave it alone. Run it once a month.

The Prompt
I want to find the repeating work in my week and decide what to do with each pattern. I'll give you 10 tasks or decisions I've handled more than twice in the last 30 days. For each one, I'll tell you what triggers it, what output I need, how long it takes, and whether I've tried AI for it before.

For each task, give me a verdict: BUILD IT (worth turning into a saved prompt, a skill, or a scheduled agent), TRY IT (worth one good manual AI experiment before automating), or SKIP IT (judgment-heavy or relationship-bound, leave it human). For every BUILD IT item, draft the actual prompt I should save and describe the trigger that should fire it.

Return as a table with columns: Task | Trigger | Output | Frequency | Time Cost | AI Tried (Y/N) | Verdict | Saved Prompt Draft.

Here are my 10 tasks:
[paste yours]

Where the BUILD IT items go: Cowork. The point of that verdict is to move a recurring pattern off your weekly to-do list and into a saved scheduled task you set once. Drop the prompt and trigger from the table into Cowork as a recurring task. (ChatGPT users can do the same with Codex Automations.)

I'm a fluent Spanish speaker. I learned the language in Chile, which means I picked up a notoriously unique version of it, full of dropped consonants, regional slang, and rhythms that don't always travel. Day-to-day conversation has never been the problem. The gaps showed up in academic register, philosophical conversations with friends abroad, industry vocabulary I never had to use in a Santiago café. For years I'd hit a ceiling and just translate around it.

AI has helped me up my game from fluent everyday conversationalist to communicating in elevated and academic register. Not as a translator (I've never really needed one), but as a continued learning tool. I feed it the rough version of what I'm trying to say, ask for the more advanced phrasing, and it gives me back not just the right words but the framing native speakers actually use. Over months, that vocabulary sticks. AI hasn't replaced my Spanish, it's deepening it.

What surprised me more was how well it handles Chilean slang. Chilean Spanish is the version everyone warns you about. The model recognizes it, responds in kind, and switches back to neutral Spanish when I'm writing to someone outside Chile. That kind of local fluency wasn't something I expected to find in a general-purpose AI, and it changes what I trust it for.

Most of us already know AI is great for business efficiency and outsourcing manual tasks. Don't forget it can also be a learning tool that fills your gaps, as long as you chat with it with purpose. Be transparent about your intentions. Tell the model you want to learn something, not just get an answer, and the back-and-forth shifts. You stop walking away with the result and start walking away with the skill.

Strategy decks pile up. Pilots run forever. Real workflows never change. We help mid-market companies skip the deck-building and get to actual implementation, starting with the two or three places AI actually moves the needle. Let's chat.