Hi {{first_name}}!
Two of the most important AI tools right now are coming from the two frontier labs most business owners are already watching: OpenAI and Anthropic.
OpenAI has Codex. Anthropic has Claude Cowork.
They overlap, but they are not the same.
This week I’m breaking down how they compare, where each one shines, and how to think about using them in real business workflows.
We’ll also cover:
OpenAI's new default model hallucinating 52% less
Apple settling a $250M lawsuit over AI features that didn't exist yet
The morning dashboard I built that replaced ten tabs and saved me an hour a day
OK, let's get into it!
New and Noteworthy
Microsoft Copilot agents are expanding to mobile. Microsoft expanded its Copilot agent capabilities across iOS and Android this week, meaning the AI workflows you build in Teams, Outlook, and other Microsoft apps can now run and surface on your phone without needing to be at a desktop. For business owners whose teams are constantly moving between devices, this closes a real gap in how autonomous agents fit into the workday.
OpenAI this week rolled out GPT-5.5 Instant as the new default model for all ChatGPT users, including free accounts. This is separate from the full GPT-5.5 released last month. Instant is the everyday, fast-response version most people interact with whether they realize it or not. The practical improvements: 52.5% fewer hallucinated claims on high-stakes prompts in medicine, law, and finance, 37.3% fewer factual errors on conversations users had previously flagged, and responses that are 30% shorter without losing substance. OpenAI also cut back on the emoji overload. A new “memory sources” feature now shows you exactly what context ChatGPT is pulling from your past chats and lets you delete or correct anything outdated. If you use ChatGPT regularly for client-facing work or research, the accuracy gains alone are a meaningful update.
Anthropic launched 10 ready-to-run finance agents. These new Claude agents are built for workflows like pitchbooks, credit memos, KYC reviews, valuation analysis, and month-end close. Even if you are not in financial services, this is worth watching. It shows where AI is heading: away from generic chat and toward purpose-built agents for specific professional workflows.
OpenAI is reportedly building an AI-native phone. Reports surfaced this week that OpenAI may be developing its own smartphone, designed around an AI agent that works continuously in the background rather than waiting to be opened like an app. No release timeline has been confirmed, but the direction is clear: OpenAI wants AI to move from something you use to something that runs alongside you throughout the day. If it ships, it could fundamentally change what “using AI” looks like for most people.
Perplexity launched an AI platform built specifically for finance professionals. Perplexity is known for AI search, but this new product pushes further into professional workflow territory with live data, document analysis, research, and structured outputs for finance teams. The bigger signal is that AI tools are becoming more vertical. Instead of one general tool for everyone, we are starting to see tools built around the way specific industries actually work.
OpenAI quietly pushed a major set of upgrades to Codex over the last couple of weeks. Codex is starting to feel a lot less like a coding tool and a lot more like an actual AI work environment. Codex can now run multiple agents in parallel, schedule recurring tasks, remember preferences across sessions, operate desktop apps on macOS, and connect into tools like Microsoft apps, Gmail, Google Drive, GitHub, and more. OpenAI is also expanding “Skills,” which let you save reusable workflows and instructions once instead of rebuilding context every time. My biggest takeaway after testing it: OpenAI is clearly moving beyond chat and trying to turn Codex into a true agentic operating layer for work.
The U.S. government now gets to inspect frontier AI models before they're released. Google DeepMind, Microsoft, and xAI agreed this week to give the U.S. government early access to unreleased AI models for national security testing, coordinated through the Commerce Department’s Center for AI Standards and Innovation. More than 40 evaluations have already been completed, including on models that never reached the public. This is a meaningful shift. The labs are no longer shipping first and letting regulators react. The bigger question is whether pre-release access eventually becomes a soft approval requirement, which would make frontier AI look a lot more like regulated infrastructure than consumer software.
Apple agreed to pay $250M over misleading AI marketing claims. A U.S. class action settled this week over Apple advertising Siri and Apple Intelligence features that had not actually shipped yet. The settlement covers roughly 37 million U.S. devices, with individual payouts depending on the model. Final approval is scheduled for June 17. The practical takeaway is simple: AI promises are easy to make and harder to deliver. If you are evaluating software for your business, test the actual workflow before buying into the marketing.
Cowork now has Competition
Here’s something I often say when asked about which frontier model people should use: ChatGPT still knows me better than any other AI tool out there. Years of history will do that. My writing style, my clients, my ideas, the way I think through problems, it’s all in there. We’ve talked about migrating your chat history to other tools but walking away from that context isn’t easy so I still find myself coming back to ChatGPT as the default (across business and personal).
That said, Claude Cowork (which is only accessible via the Claude desktop app) is something I use every single day throughout the day and genuinely recommend without hesitation. It’s a different kind of tool. Not just a chatbot replacement, it’s an AI teammate that works alongside you inside your computer.
Until recently, nothing else really felt close (unless you’re on the bleeding edge experimenting with OpenClaw and Hermes agents).
OpenAI’s recent Codex upgrades are the first time I’ve genuinely felt ChatGPT moving beyond “chatbot” and into “work operating system” territory. ChatGPT is calling it their “Superapp” and I agree it’s headed that direction.
The biggest shift is that OpenAI is starting to combine:
• persistent context
• connected tools
• reusable workflows
• scheduled automations
• desktop and browser control
• agent-like task execution
…in one environment.
Two things stand out most as I start using Codex more and more.
The first is Skills. You can save reusable instructions and workflows once, then call them up instantly later without rebuilding context every time.
The second is scheduled automations. You can now automate recurring tasks like inbox triage, summaries, reminders, workflow checks, and other repeatable tasks on a schedule.
That sounds small, but it’s actually a major shift. We’re moving from “prompting tools” toward systems that proactively help run work in the background.
Here’s my current take after spending time with both:
Claude Cowork still feels stronger for deep thinking, ongoing context, research, writing, messy workflows, dashboards, and multi-step business tasks that require judgment.
Codex feels increasingly strong for execution, repeatable workflows, technical projects, automations, internal app building, integrations, and structured task management.
In simple terms:
• Claude Cowork feels more like an AI teammate
• Codex feels more like an AI workbench
For most business owners, this is not really about which model is “smarter.” It’s about which tool fits the job.
If I’m working through strategy, organizing information, building a presentation, cleaning up a messy workflow, or creating something from scattered context, I’m usually reaching for Claude Cowork.
If I’m building repeatable systems, documenting processes, improving workflows, or setting up automations, Codex is becoming much more interesting.
The important thing to understand is that both tools are moving beyond chat.
They don’t just answer questions anymore. They can take a goal, gather context, use tools, execute steps, and return a finished deliverable.
Both tools will continue to evolve fast and both tools are competing to be your mission control layer for work, that is connected to everything, and can do everything. I’ll continue using both for now, and honestly I think there’s real value in spending time with each because they excel at different types of work.
I Built a Dashboard That Reads My Day Before I Do
Every morning used to start the same way. Open Gmail. Switch to Calendar. Check Slack. Check ClickUp tasks. Open GoHighLevel. By the time I had a clear picture of the day, 60 minutes were gone and I already felt behind.
I don't think that's a focus problem. I think it's a systems problem.
So I built something to fix it. Inside Claude Cowork, there's a feature called Live Artifacts. It's essentially a personal dashboard you build once through a conversation with Claude, and it automatically re-pulls your data every time you open it. Calendar, email, Slack, Project management tools, CRM, all in one place. You open it, scan it in 5 minutes, and start your day.
The setup in practice:
You connect 2 to 4 tools you check every morning anyway: Google Calendar, Gmail, Slack, whatever is already part of your routine. Claude walks you through a short set of questions about what you want to see, then builds a first version. No coding, no templates, just a conversation. Takes about 20 minutes total.
I'll be honest: the first version it generates is usually decent, not great. The real work is in the refinement. Once the initial dashboard is up, you give Claude your actual priority rules. What counts as urgent. What can wait. What you never want cluttering the top. That's the step I can see most people skipping, and it's the one that makes the difference between a dashboard that's interesting and one that's genuinely useful.
Mine pulls my meetings for the day, flags anything that needs prep, surfaces the Slack messages actually waiting on me, and shows a "while you were away" summary of anything that came in overnight. I open one tab. I know what matters. I start work.
A few things to know before you build yours. This requires a paid Claude plan, Pro or higher. It runs on the Claude Desktop app for Mac or Windows, not the web app, and it's machine-specific, so it won't sync across devices. Build it on the computer you use most. It also refreshes when you open it rather than streaming data in real time, which is fine for a morning briefing but worth knowing upfront.
It's not a perfect system. Some mornings the intel on a meeting is thin if there's not much context in my connected tools. And if you're on a team plan, your org admin may need to enable connectors before you can access them.
But for a busy morning? One tab instead of ten. That trade took me 20 minutes to set up, and I've gotten time back every day since. So this morning I tried to be cognizant of that time saved and spent an extra hour with the kids before school! ;)
If you're trying to figure out where to start, or you've already started and hit a wall, hit reply. I’m always happy to share guidance or recommendations based on your current situation.
What do you want me to do a deep dive on next week? Let me know. I'd love to hear.
Cheers,
Julien
PS: And if this edition was useful, please feel free to consider passing it along to someone who'd get something out of it. www.ampra.ai/join-our-newsletter.