November's practical AI question was becoming clearer: how does an assistant reach the information a business actually uses? A polished answer is of limited value if it misses the current procedure, the right customer file, or the table containing the price. Three developments addressed that gap from different directions. They offered useful starting points for controlled experiments, while leaving companies responsible for permissions, document quality, and the decisions made from the output.
1. A common connection standard arrives
On November 25, Anthropic released the Model Context Protocol, or MCP, as an open standard for connecting AI applications with data sources and tools. The initial release included a specification, developer kits, local Claude Desktop support, and example servers. Its central proposition was reducing the need to create an entirely different integration for every combination of assistant and business system. Remote production tooling was still described as forthcoming.
For an operations team, the important possibility was a more maintainable route between an assistant and approved company knowledge. That did not mean every database should immediately become available to every employee. Connecting a system makes existing access decisions more consequential. A pilot could expose an approved procedures folder and ask the assistant to answer questions with links back to the source. Useful acceptance cases would include an outdated document, two conflicting versions, and a question the repository cannot answer.
Business takeaway: Start with one read-only information source and a named data owner. Track correct answers, missing answers, and permission failures before expanding the connection.
Source: https://www.anthropic.com/news/model-context-protocol
2. SharePoint agents narrow the knowledge search
At Ignite on November 19, Microsoft distinguished between features customers could use and capabilities still being previewed. SharePoint agents became generally available, while Copilot Actions and the Employee Self-Service Agent were in private preview. The distinction mattered: a business could evaluate a focused SharePoint assistant without assuming all the broader automation demonstrations were ready for general deployment.
A narrowly scoped agent suggested a useful starting point for onboarding, service procedures, and project handovers. Instead of asking an assistant to understand the whole company, a manager could define which approved material should answer a particular class of questions. The commercial test was whether staff found a correct answer faster, including time spent checking it. A friendly response that directs someone to an obsolete policy creates rework rather than savings. Content ownership therefore belonged in the same project plan as licensing and configuration.
Business takeaway: Choose one frequently repeated internal question category. Clean up its source documents, preserve access boundaries, and compare the agent with the existing search process.
3. Tables inside documents become easier to query
Amazon announced on November 22 that Q Business supported answers from tables embedded in documents. The improvement addressed a familiar weakness in document workflows: the useful business information is often in rows and columns rather than narrative paragraphs. A purchasing guide, product specification, or operating report can lose meaning if an answer separates a number from its heading or unit.
This was a released capability, but a release announcement did not establish that a particular company's documents would be interpreted correctly. A sensible evaluation would include merged cells, footnotes, repeated column headings, and tables spanning pages. Ask a reviewer to check the returned answer against the actual row, not merely whether the response sounds plausible. For commercial work, testing an incorrect price or excluded service is more informative than repeatedly asking easy questions with obvious answers.
Business takeaway: Build a small test pack from real, approved tables. Require source verification for amounts, dates, units, and exceptions before using an answer in a quote or customer commitment.
The business opportunity was better use of information already paid for and already collected. Before buying a broader transformation, identify where staff repeatedly search, copy, or reconcile the same facts. That creates a concrete project scope and an honest baseline for judging whether a connected assistant earns its place.
Present-day invitation: Book a conversation with Brice Prather at https://loyti.com/book to explore AI consulting, custom software, and business automation for your organization. Bring one operational bottleneck, the tools involved, and the business outcome you want to improve.
Brice Prather President, CEO & Founder Life On Your Terms, Inc. (LOYTI) Turning AI and custom software into smarter operations and business growth.