AI Gets A Sales Team | Wyecliff Weekly

On Monday, Anthropic announced a $1.5 billion joint venture with Blackstone, Goldman Sachs, Hellman & Friedman, and General Atlantic to launch a new firm focused on getting Claude into mid-sized companies. The target is Blackstone's portfolio first, which alone covers more than 250 mid-sized companies, and then expansion to other private-equity-owned businesses. Blackstone President Jon Gray said the new firm is meant to break what he called the biggest bottleneck in enterprise AI right now, which is the shortage of engineers who can actually implement advanced AI inside an existing business.

A study published in Science on Sunday, led by researchers at Harvard Medical School, Beth Israel Deaconess Medical Center, and Stanford, ran 76 real emergency room cases through OpenAI's o1 reasoning model and compared the results to two attending physicians. The model offered the exact or very close diagnosis 67% of the time. The two doctors hit that mark 50% and 55% of the time. The study also tested whether the model could recommend appropriate diagnostic tests and triage decisions, and found it kept pace with or outperformed the doctors there too. The authors were careful with the framing. The two doctors were internal medicine attendings, not ER specialists. The cases were text-based, not real patients in front of a clinician. And the paper called for controlled trials before any of this gets used in actual treatment. One of the authors told the press: "We are already at the ceiling on what we can learn from these comparison studies. We need to see how this works in real care."

On Tuesday, OpenAI rolled out GPT-5.5 Instant as the new default model in ChatGPT, replacing GPT-5.3. According to OpenAI's own numbers, the new default produces 52.5% fewer made-up claims on high-stakes prompts in medicine, law, and finance, and 37.3% fewer factual errors on the kinds of conversations users had previously flagged as wrong. The response style is also tighter, with fewer filler emojis and less padding around answers. The bigger change is a new feature called memory sources. When ChatGPT uses something it remembers about you to shape an answer, you can now click to see exactly which saved memory was used, and edit or delete that memory directly. The setting is in the same Personalization menu where memory has lived for the past year.

At its annual Think conference on Tuesday, IBM announced general availability of Bob, an AI development partner the company has been using internally since June 2025. Bob started inside IBM with 100 developers and grew to more than 80,000 IBM employees in less than a year. Self-reported productivity gains across that group averaged 45%. The headline customer example, Blue Pearl, used Bob to complete a Java upgrade in three days that normally takes about 30. Bob is built to handle more than code generation. It plans, codes, tests, deploys, and modernizes legacy systems, with security and compliance controls layered on top. It works across multiple AI models, not just IBM's own. A 30-day free trial is open to any company. Pricing for the paid plans was not published with the launch.