November’s workplace AI announcements offered a tempting proposition: employees could ask questions, summarize complex material and navigate company knowledge through a conversational interface. The harder question for a business owner was where that interface would create dependable value. A useful starting point was a repeated, expensive task with an identifiable owner and a clear definition of a correct result. This edition examines three developments through that lens. Availability statements describe the position during November 2023; the recommendations are editorial analysis, not claims that LOYTI deployed these products or measured customer results.
1. Microsoft’s copilot becomes something businesses can customize
Microsoft introduced Copilot Studio on November 15, following the November 1 enterprise general availability of Copilot for Microsoft 365. The new low-code product was presented as a way to customize copilots and build standalone assistants. Microsoft also emphasized that Microsoft 365 Copilot operated within a user’s existing identity and permissions. That matters because an assistant can make previously obscure information much easier to find, even when it has not technically expanded someone’s access.
For a business evaluating this direction, permission cleanup belongs beside the pilot plan. A proposal-writing assistant should draw from approved service descriptions, current pricing rules and relevant examples. It should not discover an obsolete price sheet and confidently make it easier to reuse. Give the pilot a bounded assignment, such as turning an approved meeting record into a draft follow-up. Have its owner check names, commitments and next actions against the original. Measure both drafting time and correction time.
Business takeaway: start with one document workflow and its information permissions. A successful demonstration is a reason to test a repeatable process, not a reason to license every employee immediately.
2. Claude 2.1 makes larger document reviews possible
Anthropic released Claude 2.1 with a 200,000-token context window and introduced tool use in beta. A larger context window increased how much material could be supplied for a task. Anthropic also reported improvements in its own evaluations. Those findings were vendor evidence; they did not establish that every long document would be interpreted correctly or that important details could no longer be missed.
The practical opportunity was a first-pass review of substantial manuals, procurement packets or policy collections. Instead of asking for a general summary, define the questions that matter: What obligations require action? Which dates conflict? Where does the document explain an exception? Require section references and separate direct evidence from interpretation. Keep a list of questions whose answers are deliberately absent from the material. A useful assistant should identify missing information instead of filling the gap with a plausible answer. Sensitive documents still require an approved handling process.
Business takeaway: longer context can reduce document preparation work, but source checking remains part of the job. Evaluate omissions as carefully as incorrect statements.
Source: https://www.anthropic.com/news/claude-2-1
3. Amazon Q brings internal knowledge into the conversation
AWS announced Amazon Q on November 28. The broader assistant was introduced in preview, with enterprise information connections and access controls among its proposed features. The announcement described more than 40 connectors. That was a useful indication of integration ambition, but a connector count is not evidence that a particular company’s documents are accurate, consistently labeled or ready for reliable answers.
Consider the recurring questions that interrupt a small operations team: Where is the current onboarding checklist? Who approves a refund? Which instructions apply to this service package? Before connecting everything, collect a sample of those questions and identify the authoritative answer for each. Give documents an owner and an update date. Test the assistant as users with different permissions, and check that restricted material stays restricted. If two policies disagree, resolve the underlying disagreement rather than merely rewriting the prompt.
Business takeaway: an internal assistant is most valuable when it points people to maintained operating knowledge. Organizing that knowledge is productive work even if a preview does not make it into production.
Source: https://press.aboutamazon.com/2023/11/aws-announces-amazon-q-to-reimagine-the-future-of-work
A practical next step
Choose ten real questions or documents, record how long the existing process takes, and score the AI-assisted result for accuracy, missing information and effort to correct. Keep that baseline when changing tools. It turns a product demonstration into a business decision.
Present-day invitation: Have a workflow that consumes time or slows customer service? Book an appointment at https://loyti.com/book to discuss AI consulting, custom software and business automation with Brice Prather, President, CEO & Founder of Life On Your Terms, Inc. (LOYTI). Bring the process and the problem; the conversation can focus on what an improvement would need to achieve.