July's AI news made measurement harder and more important at the same time. Agents were taking on longer tasks, research examined adoption at organizational scale, and vendors reported increasing use. More activity could signal progress, but it could also mean more output to review, more compute to pay for, or more work waiting downstream. For a business leader, the useful measure was a completed outcome that passed review and improved the operation, with its full cost included.
1. Coding-agent research separates output from value
A paper submitted July 1 studied Microsoft's early-2026 rollout of command-line coding agents across tens of thousands of engineers. Its authors reported that peer networks influenced initial adoption and that adopters merged roughly 24 percent more pull requests than their estimated counterfactual. They explicitly treated merged pull requests as an output proxy, not the value delivered. This was research in a particular organizational setting, not a promise that any development team would gain the same percentage.
Business takeaway: Judge a software project by useful changes reaching customers reliably. Pair delivery speed with defects, review time, maintenance burden, and user adoption. Encourage a small group of capable users to share practical examples, including failures, rather than relying only on a general training session. If code generation increases faster than review capacity, invest in better tests and clearer requirements. More changes are an advantage only when the business can safely absorb them.
Source: https://arxiv.org/abs/2607.01418
2. ChatGPT Work introduces longer task execution
OpenAI's July 9 Business release notes introduced ChatGPT Work for longer tasks involving research, connected applications, files, and finished deliverables. The release described progress visibility, questions, direction changes, and approval of important actions. Availability rolled out by plan, and certain workspace settings and preview periods applied. This was a release announcement with specific access conditions, not evidence that every business workflow could be handed over without setup or review.
Business takeaway: Start with an internal deliverable that has a clear quality bar, such as a source-linked competitor briefing or a draft operating manual. Identify which connected information the task actually needs. Tell the system what it may produce and what it may not publish, send, or change. Keep the reviewer responsible for factual accuracy and commercial context. A finished-looking document is still a draft until someone with the right knowledge has checked the claims and approved its use.
Source: https://help.openai.com/en/articles/11391654-chatgpt-business-release-notes
3. Seat growth is a signal, not your return
Microsoft's July 30 article reported more than 30 million paid Copilot seats and argued that the next measure of progress should be transformed work. Its evidence included company telemetry and customer examples. These are useful indicators of adoption within Microsoft's ecosystem, but a paid seat is not a completed task, and aggregate usage cannot establish the return on a particular customer's implementation.
Business takeaway: Create a small operating scorecard before expanding a deployment. Track the number of tasks completed, the proportion accepted without substantial rework, time spent on review, and total cost. Add a business metric such as proposal turnaround or time to resolve a customer issue. Review the scorecard with the people doing the work. Cancel unused access or redesign low-value tasks rather than treating rising usage as the target. The goal is better service or capacity, not maximizing interaction with software.
Put it to work
Pick one recurring deliverable and follow it all the way to acceptance for two reporting cycles. Record who prepared it, who checked it, what changed, and whether anyone used it to make a decision. If AI shifts effort from the author to the reviewer, make that visible. Use a brief post-task review to improve the template and source collection. Only then decide whether more licenses, a custom integration, or a different process would produce the next useful improvement.
Present-day invitation: Turn a recurring business bottleneck into a practical improvement. Book an appointment with Brice Prather at https://loyti.com/book to discuss AI consulting, custom software, and business automation. Bring a workflow, a business goal, and the questions holding you back.
Brice Prather President, CEO & Founder Life On Your Terms, Inc. (LOYTI) Turning AI and custom software into smarter operations and business growth.