I was using machine learning to score sales leads long before ChatGPT. I took past deals that we won. I looked at what those customers had in common. Then I scored new leads against that list.

I published that method in 2022 in Applied Marketing Analytics (Bellamkonda, 2022). After ChatGPT, most people think AI means a chatbot that writes email. The older use is still useful. You can study what already worked and guess who will buy next.

Microsoft, Salesforce, and ServiceNow put these chat tools inside software you already buy. When employees still preferred Claude, the vendors added Claude as well.

Three Things People Call AI

When a vendor says AI, they usually mean one of three jobs.

The first is analysis. You look at deals you won and deals you lost. You ask which facts showed up together. Job title. Company size. How many meetings before a proposal. Which emails got a reply. Sales often ignores a marketing list. A model that only reads history has no side to take (Bellamkonda, 2022).

The second is prediction. You take a new lead and give it a score. The score is how close that lead looks to past wins. A sales rep can start Monday with twenty names instead of a full list.

The third is writing. Copilot, Claude, Amazon Quick, and Salesforce Agentforce draft the email, the slide deck, and the call notes. They are good at that. They will not learn from your won deals unless you give them that file and keep score somewhere else.

Salesforce's new Claude tools can read live fields in the CRM, the system that stores accounts and pipeline. That helps a manager see if a deal is healthy. It is still not the same as a model trained on your own closed business.

What Developers Actually Use

GitHub Copilot began as an OpenAI tool inside Microsoft. Developers can now pick OpenAI, Anthropic Claude, Google Gemini, or Microsoft's own coding model (InfoWorld, 2026).

Cursor is a code editor many teams standardized on. You can switch between Claude, OpenAI models, Gemini, and Grok while you work. SpaceX bought Cursor's parent company on August 14. OpenAI said it will stop selling its models directly to Cursor on November 12 (shashi.co, 2026). Until that date, OpenAI models still work in Cursor.

Claude Code stays on Claude. Developers use it when they want the tool to work through a whole code project. OpenAI Codex stays on OpenAI. Replit and Windsurf, now part of Cognition's Devin Desktop, use the same short list of big models and add one of their own.

Why Companies Turned On Copilot First

Most large companies already pay for Microsoft 365. Turning on Copilot was the easy switch. Questions and files stayed inside the company's Microsoft account. Security teams could say yes without sending documents to a public chat site.

The first version did not feel like ChatGPT. People opened Claude. They also kept a personal ChatGPT window. Reco looked at companies on its security platform in 2026 and found four in five AI tools had no IT approval. That rate was worse than unapproved ordinary software in the same study (Reco, 2026).

Many companies then bought Claude on purpose, next to Copilot. Ramp's May index said 34.4 percent of the businesses it tracks pay Anthropic, and 32.3 percent pay OpenAI. Those two numbers add to more than 100 percent of AI buyers in the sample. A lot of those companies pay both (Ramp, 2026).

Copilot now works in Word, Excel, Outlook, and Teams as one assistant. Microsoft said on July 30 that Microsoft 365 Copilot passed 30 million paid seats (Microsoft, 2026).

Mustafa Suleyman's Microsoft AI group released its own MAI models in June. GitHub Copilot now lists MAI-Code next to OpenAI, Claude, and Gemini (GitHub, 2026). Hard questions inside Microsoft 365 still go to OpenAI and Claude. Suleyman told The Verge, "There's three labs that matter, Google DeepMind, OpenAI, and Anthropic. We are not one of them at the moment." He said Microsoft has to prove it can train models from scratch (The Verge, 2026).

Salesforce and ServiceNow

Salesforce spent two years putting OpenAI, Claude, Gemini, and its own models behind Einstein and Agentforce. On August 26 it named the Claude pairing Claudeforce. Claude is now the default in Slack. It also runs Agentforce Vibes and Agentforce Coworker unless you change it.

Salesforce in Claude is a plugin with 37 ready-made sales tasks. It is in pilot now. An open beta is expected in September (Salesforce, 2026). Agentforce already shows up inside ChatGPT from a separate OpenAI deal announced at Dreamforce in October 2025 (Salesforce, 2025).

ServiceNow lets Now Assist run on ServiceNow's own model, Microsoft's Azure OpenAI, Google Gemini, or Claude hosted on Amazon. OpenAI is the usual choice for voice and images under a January 2026 deal. Claude is the default for Build Agent, the tool that lets you describe a workflow in plain English (OpenAI, 2026; ServiceNow, 2026).

Amazon Quick and Nova

Amazon Quick is Amazon's work assistant. Like Gemini, Claude, and Copilot, it can connect to Microsoft 365, Google Workspace, Slack, Salesforce, Zoom, and files on your computer. You can ask it for a report, a deck, a ticket, or a task that runs in the background. DXC rolled it out to 115,000 employees in 70 countries (Amazon Web Services, 2026; DXC, 2026).

You do not have to move your work onto Amazon's cloud first. Quick plugs into the tools you already use. Amazon Q Developer is the separate coding tool for software teams.

Nova is a different product. It is Amazon's own family of models that run on Amazon Bedrock, the service where AWS customers rent models. Nova 2 Lite is the everyday model. Nova 2 Sonic handles speech. Nova Premier, Canvas, and Reel are being retired. Premier is listed to shut down on September 14, 2026 (Amazon Web Services, 2026). Amazon also sells Claude on Bedrock and is a major investor in Anthropic.

Grok

Grok from xAI is now in the same SpaceX group as Cursor. I find it fast and often accurate. A general counsel still has to approve the vendor. That review is harder when the owner, the social network, the rocket company, the code editor, and the model lab share one public voice.

Grok 5 was expected in late 2025, then in the first quarter of 2026, then in the second quarter. It has not shipped. Grok Business and Grok Enterprise exist. Enterprise Technology Research surveyed about 500 people using March 2026 data. Claude was in use at 48 percent of those companies, Gemini at 40 percent, and Grok at 7 percent (Enterprise Technology Research, 2026).

Good answers are not the same as getting on an approved-vendor list. Missed ship dates slow that review down.

What Happens Next

Watch whether Microsoft's own MAI models handle enough Copilot work that Microsoft pays OpenAI and Anthropic less for everyday questions.

Amazon is expected to show a new big model around its re:Invent conference. That will tell you if buyers start asking for Nova by name, or if they keep using Quick as the assistant and Claude as the model underneath.

When you write next year's approved-tool list, count how many personal chat windows still sit outside it.

About Shashi Bellamkonda

Loading bio…

LinkedIn Profile

Get new posts by email:

Subscribe

Powered by