The GPT-5.4 update is a massive leap forward — if you're willing to look past the controversy.

Until recently, I hadn't been logging into ChatGPT much. If I am being honest, it felt like it had fallen behind. The outputs were getting predictable, the reasoning wasn't matching what I was seeing elsewhere, and in the great game of LLM leapfrog, Claude had taken a commanding lead for my daily workflows.

Then last week happened. OpenAI dropped GPT-5.4, a major upgrade that fundamentally changes how the tool works. But the launch was entirely overshadowed by the news that OpenAI had signed a major contract with the U.S. Department of Defense.

The backlash was immediate. Over 1.5 million users pledged to boycott the app, uninstalls of the mobile app surged by nearly 300%, and several senior employees resigned in protest. Anthropic's Claude became the #1 free app in the App Store as users fled.

It is a fascinating paradox: just as millions of people are abandoning ChatGPT on principle, the tool itself has become significantly better for professional work. I am still not sure if I like the new ChatGPT more than Claude overall. But after spending time with GPT-5.4 this week — including using Agent Mode to do something I had never tried before — I think it is worth giving it another look.

AI LESSON

Five things you can do today with the new ChatGPT that you couldn't do before.

The GPT-5.4 release comes with a lot of noise about benchmarks and token efficiency. But for business professionals, five updates stand out as immediately actionable: the ability to interrupt the AI while it's thinking, a new native Excel integration, Agent Mode's autonomous web browsing, a surprisingly competitive Deep Research tool, and the ability to integrate ChatGPT directly into the apps where you already work. Each of these can be put to work the same day you read this.

GPT-5.4 Thinking is available on Plus ($20/mo), Pro ($200/mo), and Business/Enterprise plans. Free users get access to the updated GPT-5.3 model but not GPT-5.4. To switch, use the model picker at the top left of your chat window and select "GPT-5.4 Thinking".

What it does: GPT-5.4 Thinking shows you an upfront plan of its reasoning before it writes the full response. You can interrupt and redirect it while it is still thinking, rather than waiting for a finished output and starting over.

Step 1: Select GPT-5.4 Thinking from the model dropdown.

Step 2: Enter a complex, multi-step prompt — for example, asking it to draft a strategic marketing plan based on three different audience personas.

Step 3: Watch as ChatGPT outlines its "Thinking Plan" before writing the final output.

Step 4: If you see the plan going in the wrong direction, click the Interrupt button. Type your correction — "Actually, focus entirely on the enterprise persona and ignore the SMBs" — and the model will pivot immediately.

A practical example: Paste a long vendor agreement and ask it to identify risk areas. When it starts listing standard boilerplate risks in its plan, interrupt it and say, "Skip the standard indemnity clauses and only look for non-standard data privacy risks." You get exactly what you need without waiting for a useless first draft.

Limitation: GPT-5.4 Thinking messages are capped (up to 3,000 per week for Plus users), so save this for complex, high-stakes tasks.

What it does: The new ChatGPT for Excel add-in embeds ChatGPT directly inside Microsoft Excel, letting you build, update, and analyze live spreadsheet models using natural language.

Step 1: In Excel, go to Home → Add-ins, search for "ChatGPT", and install the official add-in.

Step 2: Sign in with your ChatGPT account.

Step 3: Open a raw data sheet and type your request in the ChatGPT side panel:

Step 4: ChatGPT explains the formulas it intends to use. Click to approve, and it builds the model directly in your workbook using native Excel formulas.

A practical example: If you have inherited a messy spreadsheet from a former colleague, ask the add-in to "Explain how the assumptions in Sheet 1 flow into the final revenue projection in Sheet 3, and highlight any hardcoded numbers that should be formulas." What used to take 45 minutes of manual tracing takes about three minutes.

Limitation: The add-in is currently in beta and works only in Microsoft Excel. Google Sheets support is not yet available.

This is the use case that actually brought me back to ChatGPT.

Agent Mode — accessed via the tools menu (the "+" icon) in the chat composer — lets ChatGPT browse the web autonomously across multiple pages to complete a research task. It doesn't just search for and summarize a single result. It opens pages, reads them, follows links, and filters what it finds based on your specific context. You can also connect it to external data sources using Apps (found under Settings → Apps in the left sidebar), allowing Agent Mode to search your Google Drive, Slack, or Notion.

I tried this prompt: "Look for all the most interesting new products listed on Product Hunt this month and then find ones that best fit my workflow."

What happened next was genuinely impressive. ChatGPT searched for the best products of March 2026, opened the Product Hunt results page, navigated to the aggregator site hunted.space to get daily launch lists, then opened individual product pages one by one — Chronicle, Macaly, Spine, Sonarly, GapHunt, Agent Skills, and more — reading each one and summarizing its features in the context of my workflow. It filtered a list of dozens of new tools down to the handful most relevant to what I actually do.

Step 1: In ChatGPT, click the "+" icon in the chat box and select "Agent mode" from the dropdown. (Optional: To let it search your internal files, go to Settings → Apps in the left sidebar to connect your Google Drive or Microsoft 365).

Step 2: Give it a research task with a clear personal context. The more specific you are about your role and workflow, the better the filtering:

Step 3: Watch as it browses autonomously. You can monitor its progress in the activity panel. It will show you which pages it is opening and what it is reading.

Step 4: When it finishes, review the curated list. You can ask follow-up questions — "Tell me more about Chronicle" or "Which of these have a free tier?" — and it will continue browsing to answer.

A practical example: Use this before your weekly planning session. Ask Agent Mode to scan Product Hunt, relevant subreddits, or industry news sites and surface the three tools most worth evaluating this week. It replaces 30–60 minutes of manual browsing with a five-minute review.

Limitation: Agent Mode can occasionally get stuck on pages that require a login or block automated access. It will tell you when it hits a wall, but it does not always find a workaround on its own.

The biggest bottleneck in AI workflows is moving data back and forth between your tools and the chat window. OpenAI has quietly solved this by renaming their "Connectors" to Apps and making them much more powerful. But that’s not the only helpful but confusing because there’s also a huge app library on the left fo the screen.

What it does: Apps let you securely connect external services directly to ChatGPT. Instead of copy-pasting documents, you can @-mention your Google Drive, search your Slack history, or have ChatGPT take actions in other platforms right from the chat interface.

Step 1: In the ChatGPT left sidebar, click Settings → Apps to open the App Directory.

Step 2: Browse the available apps and click Connect on the tools you use daily (like Google Drive, Slack, or Notion).

Step 3: Complete the authentication flow.

Step 4: In any chat, simply type @ followed by the app name (e.g., @Google Drive) to search your connected files and pull them into the conversation as context.

A practical example: You are drafting a Q3 strategy memo. Instead of digging through your drive for last quarter's numbers, you type: "@Google Drive find the Q2 performance review deck and summarize the top three missed targets, then use that to draft the opening of my Q3 strategy." ChatGPT finds the file, reads it, and writes the draft without you ever leaving the window.

Limitation: App availability depends heavily on your plan. Business and Enterprise/Edu plans have apps enabled by default (though admins can control them), while some apps may not be available on Plus/Pro depending on the partner.

Deep Research is one of my favorite AI workflows, and until recently, Google Gemini's Deep Research was where I did it. Gemini is excellent — it generates a detailed research plan, and the output quality has been consistently strong. I had no reason to switch.

But after testing ChatGPT's version this week, I found the real advantage isn't just the research itself — it is what you can do after. Because ChatGPT combines Deep Research with Agent Mode and Apps, you can create a continuous research-to-action loop. You don't just get a report; you get a report that the AI can immediately turn into a slide deck or push into your company wiki.

What it does: Deep Research executes multi-step, complex queries over 5–30 minutes, delivering a structured, heavily cited report. But the magic happens when you chain it with other tools.

Step 1: In the ChatGPT prompt window, type /Deepresearch or select "Deep research" from the tools menu (+).

Step 2: Give it a complex question: "Research the current market rates for enterprise AI consulting services in the US, including pricing models and standard deliverables."

Step 3: Review and approve the generated research plan.

Step 4 (The Magic Step): When the report finishes, don't just download it. Tell ChatGPT: "Now switch to Agent Mode, use this research to write a 10-slide presentation outline, and use the Google Drive App to save it directly into my 'Q2 Strategy' folder."

A practical example: You need to brief your team on a new competitor. You run Deep Research on their recent product launches and pricing changes. Once the report is done, you ask ChatGPT to draft an executive summary and use the Slack App to post it directly to your #leadership-team channel. What used to be three separate tasks across three different tabs is now one continuous conversation.

Limitation: It is slow by design (taking up to 30 minutes for complex queries) and usage is capped based on your plan. The handoff between Deep Research and Agent Mode can also occasionally drop context if your follow-up prompt isn't specific enough.

GPT-5.4 is not fully autonomous. The Thinking feature still requires a human to review and approve its plan. The Excel add-in works only in Microsoft Excel for now. And the most advanced customization feature — "Skills" for creating reusable workflows — is currently locked behind Business and Enterprise tiers, leaving Plus users without it for now. The Pentagon deal also raises legitimate questions about data governance that enterprise buyers will need to evaluate carefully before expanding usage.

The fastest way to experience what's changed is to try these use cases this week. Start by switching to GPT-5.4 Thinking and giving it a complex prompt you previously abandoned. Then connect your Google Drive via the Apps menu, open Agent Mode to send it on a web browsing task you would normally do manually, and kick off a Deep Research query before you step away for lunch. Finally, install the Excel add-in to format your next reporting spreadsheet. Each one will show you something the previous version of ChatGPT simply couldn't do.

I appreciate your support.