A 60-minute walkthrough for setting up Anthropic's new desktop agent and getting your first multi-step tasks running on their own.

What we'll cover:

• What Claude Cowork actually does, including how it reads your files, works across your apps, and executes multi-step tasks on its own
• How to install, configure, and run your first agent from scratch
• Real agent examples you can adapt for your own recurring workflows

Whether you're brand new to Cowork or already trying to push past one-off setups, this session will give you the practical setup and agent-design patterns to get value out of it from day one. Note: requires a paid Claude plan.

May 14th at 10:00 AM PT / 12:00 PM CT / 1:00 PM ET
Save your spot here →

• OpenAI launched GPT-5.5 and ChatGPT Images 2.0 the same week. GPT-5.5 (April 23) ships in ChatGPT and Codex with API access April 24, leaning into agentic coding, computer use, and deeper research — plus a same-day GPT-5.5 Pro tier. ChatGPT Images 2.0 (gpt-image-2) adds a "Thinking" mode that reasons before drawing, 2K output, and consistent characters across frames. (GPT-5.5, Images 2.0)

• DeepSeek released V4-Pro and V4-Flash on April 24 as open-weight preview models with a 1-million-token context window and pricing at $0.14 / $0.28 per million input/output tokens for Flash. Coding benchmarks rival U.S. frontier labs at a fraction of the cost. Tencent and Alibaba are reportedly in talks to invest at a $20B+ valuation. (DeepSeek, TechCrunch)

• Google committed up to $40 billion to Anthropic ($10B immediate at a $350B valuation, $30B tied to milestones) plus 5 GW of Google Cloud compute over the next five years. Coming one week after Amazon's $25B commitment, Anthropic now has compute and capital lined up across all three hyperscalers. (Bloomberg, CNBC)

• Microsoft 365 Copilot now has GPT draft and Claude review, inside its Researcher agent. OpenAI's model writes the response, Anthropic's model checks accuracy and citation quality before the user sees it. GeekWire is calling it the end of the "single-model era" in enterprise AI. (Microsoft, GeekWire)

• Google Cloud Next 2026 wired Gemini across the stack. The new Gemini Enterprise Agent Platform offers 200+ models (including Anthropic's Claude), a $750M partner fund, a no-code agent builder, production-grade Agent2Agent (A2A) protocol, and managed MCP servers. Workspace Intelligence pushes Gemini deeper into every app — Gmail AI Overviews, Sheets "Fill with Gemini," prompt-to-deck in Slides, and continuous awareness across Mail, Drive, Calendar, and Docs. (Google Cloud, Google Workspace, TechCrunch)

• Anthropic published a postmortem on Claude Code quality regressions, identifying three separate bugs across March and April: a reasoning-effort default change, a thinking-cache flaw that wiped context on every turn, and a verbosity instruction that quietly cut coding quality 3%. All resolved by April 20; usage limits reset April 23. Worth knowing if your team felt Claude was "off" the past month. (Anthropic Engineering)

• Project Prometheus, Jeff Bezos's "physical AI" lab co-founded five months ago with Vik Bajaj, closed a $10 billion round at a $38 billion valuation, with JPMorgan and BlackRock participating. The lab has 120+ employees poached from OpenAI, xAI, DeepMind, Meta, Anthropic, and Nvidia, no shipped product yet, and is focused on AI for engineering and manufacturing physical goods. (Bloomberg)

Three findings worth knowing this week.

• Salesforce surveyed sales professionals and found 87% of sales organizations now use AI for tasks like prospecting, forecasting, and email drafting. Sellers using AI agents report a 33% time reduction on research and content creation, and 89% say AI is improving their understanding of customers. (Salesforce State of Sales 2026, February 2026)

• An NBER randomized experiment with 1,174 participants found AI access closed three-quarters of the productivity gap between higher- and lower-education workers. Without AI, college-educated workers significantly outperformed those without degrees. With AI, that gap nearly disappeared. (NBER Working Paper, March 2026)

• The New York Fed surveyed 1,200 U.S. workers and found only 16% report that their employer offers AI training, even though 38% say such training would be valuable. AI adoption at work is highly skewed: 58.7% of college graduates have used AI tools versus 22.9% without a degree. (NY Fed Liberty Street Economics, April 2026)

[Paste a draft you're working on, an email, memo, deck outline, plan, proposal, anything where the stakes matter. Tell me what it's for, who the audience is, and what outcome you're trying to drive. If you already have specific concerns, say so.]

Read what I gave you twice. First, as a thoughtful skeptic: tell me where the argument is weakest, where the writing is doing extra work to cover for thin reasoning, and what a smart reader would push back on. Don't soften it. Be direct about what I'd be embarrassed to defend. Then, as someone who has to actually act on what I'm saying: tell me what's missing for them to take the next step. A number, a name, a clear ask, a concrete commitment. Tell me what they'd ignore and what they'd remember. Don't rewrite the draft. Don't tweak tone. Tell me what to change about the substance.

Return it as: skeptic critique + action-taker critique + the three substantive changes that matter most.

I've spent a lot of these notes talking about how to pick the right tool from the same lab for the right job. Claude chat for steering. Cowork for background tasks. Claude Code for anything that touches files. Design for visuals. Once you stop treating every interface the same, the work gets faster.

What I'm realizing now is that the same logic applies across labs, not just within them. I've been a heavy Claude user for a while, no secret to anyone who reads these notes. But OpenAI's GPT-5.5 in Codex is a beast at certain things, and pretending it isn't because I'd rather stay inside one ecosystem leaves real value on the table.

If you have access to a few subscriptions, the move isn't to standardize on one. It's to pit them against each other or have them work in concert, depending on the task. Some weeks I'll start a problem in Claude chat to think it through, then hand the actual build off to Codex. Other times I'll have both draft the same thing and pick the one that lands.

Pay for the stack you have, and use each tool where it actually wins. If your team is on ChatGPT, Claude, and Gemini, treat that as a deliberate stack, not a redundancy problem.

Kiingo helps mid-market companies figure out what actually works with AI and stop spending time on what doesn't. Training, strategy, implementation.

If your team is using AI but you're not sure it's moving the needle, that's exactly the conversation we have. kiingo.com/contact