I Stopped Using ChatGPT. Here's Why.

Infrastructure Catalyst
Issue #5 | January 25, 2026

🎯 THE LLM RACE: WHERE WE STAND

I've been using AI tools for over three years now. Watched the models improve, the interfaces change, the companies rise and stumble. A year ago, I switched from ChatGPT to Claude. I haven't switched back.

This isn't about declaring a winner. The race is far from over. But something shifted in 2025, and the data backs it up.

Since 2023, we've been watching a battle between LLM builders unfold in real time. OpenAI, Anthropic, Google, Microsoft, Meta, xAI, and dozens of others chasing model training, each taking slightly different approaches.

ChatGPT started the race with roughly 87% market share and the first-mover advantage. Then they tried to go in every direction at once: DALL-E for images, Sora for videos, Codex for coding, and the recent GPT-5.2 model. They spread themselves thin, and it's showing. According to Similarweb data, ChatGPT's market share dropped to around 65% by January 2026, a 22-point decline in twelve months. Sam Altman issued a "code red" memo in December, redirecting the company's focus back to ChatGPT.

Google is executing. Veo3 leads video generation. Nano Banana Pro is driving image generation adoption. Vertex AI is becoming the enterprise platform for building workflows, agents, and apps. Gemini grew from 5.4% to over 20% market share in that same period, according to Similarweb's January 2026 report.

Microsoft is betting heavily on OpenAI's future while developing Copilot. It's a gamble that hasn't fully paid off yet.

Meta slowed down, betting on open-source models. Enterprise adoption of open-source actually declined from 19% to 13% in the first half of 2025, according to Menlo Ventures' mid-year report. Their Superintelligence Labs lost key engineers within months of launch, and they've since cut 600 jobs from the unit.

Elon Musk is trying to make a comeback with xAI and Grok. They've grown to 3.4% market share and lead in session duration, but they're still catching up on enterprise adoption.

Then there's Anthropic, the company behind Claude. If you haven't heard of them, Claude is the AI assistant I use daily and the subject of this newsletter.

What's astonishing about the last year is how clearly Anthropic understood what users actually need. They didn't try to do everything. They focused.

First, they owned creative writing. Claude consistently produces output that feels less robotic, more human. Benchmark after benchmark, user feedback after user feedback, Claude's writing quality stays one step ahead.

Second, they went deep on coding. "Claude Code" launched and quickly became the tool developers rely on for writing clean code. Claude now holds 54% of enterprise coding market share, more than double OpenAI's 21%.

The result: Anthropic now leads enterprise AI with 40% market share, up from 32% mid-year, overtaking OpenAI's 25%, according to Menlo Ventures.

Claude played the long game. Focus on one thing, get it right, then move to the next. They understood the core problem with LLMs, the memory limitations, and they built solutions: "Projects" for persistent knowledge, "Skills" for reusable workflows, "MCP" for tool connections.

It's still early to know who wins this race. But Anthropic's trajectory over the past 18 months has been remarkable.

And now they've released "Claude in Excel."

🛠️ DEEP DIVE: "Claude in Excel" and the Feature Timeline

For those of us in infrastructure, engineering, and construction, spreadsheets aren't optional. Budgets, cost tracking, resource loading, bid analysis, change order logs. Excel is the backbone.

"Claude in Excel" is an add-in that puts Claude directly inside your spreadsheet. Not a separate window. Not copy-paste. Claude reads your workbook, understands your formulas, and makes changes in real time. It runs on Opus 4.5.

Microsoft Copilot has shipped Agent Mode and a COPILOT function for Excel, but Microsoft themselves warn against using it for "tasks requiring accuracy or reproducibility" or anything with "legal, regulatory, or compliance implications." That's a significant caveat for AEC work where numbers matter.

Infrastructure PM use cases:

Ask Claude to explain a multi-tab budget you inherited from someone else. It walks through the structure, identifies where formulas pull from, and flags issues.

Describe a pay application tracker you need: contractor name, scheduled value, work completed to date, retainage held, amount due this period. Claude builds it with working formulas.

Point it at a spreadsheet with formula errors and ask it to find and fix them. It traces the problem faster than you can manually.

What it won't do: Conditional formatting, data validation, macros, VBA.

Security note: Don't use it with spreadsheets from untrusted sources. Prompt injection is a real risk with downloaded templates or vendor files.

⚡ WHAT CLAUDE SHIPPED (TIMELINE)

Here's what Anthropic released in the past 18 months:

"Projects" (July 2024): Persistent knowledge bases across conversations. Upload specs and standards once, reference them in every conversation. Announcement

"Claude Code" (November 2024): Terminal-based AI that reads and modifies files directly. Not just for developers. Best practices

"MCP Servers" (November 2024): Connect Claude to external tools like Notion, Slack, databases, Google Drive. Now at 97+ million monthly SDK downloads. Documentation

"Skills" (October 2025): Reusable instruction packages that load automatically. Build your own workflows for proposals, reports, or firm standards. GitHub repo

"Cowork" (January 2026): "Claude Code" for non-technical users. Point it at a folder, describe what you want. Mac only, Max subscription required. Announcement

"Claude in Excel" (January 2026): AI inside your spreadsheet. Available to all paid plans. Help article

⚡ TIPS FOR GETTING THE MOST OUT OF CLAUDE

Organize your work in "Projects," not random chats.

Create "Projects" for each major area: one for each active job, one for business development, one for professional development. Upload your key documents to the project knowledge base. Claude references them automatically.

After every significant conversation, ask Claude to summarize the key decisions and context. Save that summary to your project knowledge. This is how you build long-term memory.

Use "Skills" for your voice and standards.

If you want Claude to match your tone, formatting preferences, or firm standards, build a "Skill." It's a folder with a SKILL.md file containing your instructions. Claude loads it automatically when relevant.

Use "Claude Code" to build anything.

Despite the name, it's not just for developers. I've used it to process large datasets, batch-rename files, and build automation scripts. If you can describe what you want, "Claude Code" can probably build it.

The LLM race isn't over, and staying loyal to one tool is limiting in a market that changes this fast. That said, Claude's Max plan is the closest I've found to a generalized AI agent that fits how I actually work. It's accelerating my workflows in ways I didn't expect. For now, that's enough.

Hit reply and let me know: are you using any of these features yet?

Joseph Dib, PE, PMP
Infrastructure Catalyst