top of page

Search Results

Search results for "artificial intelligence"

256 results found

  • Artificial Intelligence, ChatGPT & Technology Impacting The World Today

    Executive Coach, Certified Organizational Development Coach, Certified Life Coach, and Certified Positive Intelligence Such technology includes: artificial intelligence and machine learning, quantum computing, renewable

  • Book Smart, Street Smart, and the Machines We Now Work With

    Most of us grew up hearing about two kinds of intelligence. That same distinction turns out to be one of the clearest ways to understand artificial intelligence. Benchmark scores, the standardized tests of the AI world, measure this kind of intelligence well, and

  • Finding Hidden Capacity in Your 2027 AI Budget

    Boston Consulting Group found that companies expected to roughly double their AI spending in 2026, from about 0.8% of revenue to 1.7%, and that 65% of CEOs ranked accelerating AI among their top three priorities (Apotheker et al., 2026). As budgets for 2027 are finalized, most leadership teams have moved past asking whether to invest and are focused on getting the most work from every dollar they commit. That focus matters because AI is billed differently from almost anything else a company buys. Much of the cost now arrives as tokens, the small units of text a model reads and writes, metered on every request. Unlike a fixed license fee, token spend grows quietly with each prompt, each document pasted in, and each new team that finds a use for the tools. The companies that pull ahead in 2027 will be the ones that manage tokens with the same care they give to cash, inventory, and their people's time. Where the Waste Hides In most organizations, AI spending is spread across departments, vendor accounts, and individual credit cards. Larridin's State of Enterprise AI 2026 research found that the average enterprise runs 23 different AI tools, yet only 38% keep a complete inventory of what is in use (Larridin, 2026). Gartner's 2027 CEO Survey points in the same direction, finding that the main obstacle to CEOs' AI ambitions is operational readiness rather than a shortage of new ideas (Gartner, 2026). This is the central argument of TokenOps: From Token Waste to Competitive Advantage, a book I wrote earlier this year. Token waste usually comes from habits that made sense during early experiments, such as using the most powerful model for every task or sending the same long instructions with every request, and that were never revisited once usage grew. Each habit is small on its own, but across thousands of requests a day they can consume a meaningful share of the budget without adding value. The encouraging news is that most of this waste can be recovered without reducing what teams accomplish. In many cases, the same changes that lower cost also make the tools faster and easier to use. Five Places to Find Capacity The following areas are where leaders most often find room in their AI budgets. Each comes with a question worth bringing to your next leadership or technology review. Match the Model to the Task AI providers offer families of models at very different prices. The most capable models earn their cost on complex analysis, strategy work, and difficult writing. Routine tasks such as sorting emails, summarizing meeting notes, or pulling figures from invoices can usually be handled by smaller models that cost a fraction as much and respond faster. Many organizations default to the top model everywhere simply because it was the one used in the pilot. Question to ask: Which of our high-volume tasks are running on our most expensive model, and do they need to be? Trim What Every Prompt Carries Every request pays for everything sent along with it, including standing instructions, background documents, and the full history of a long conversation. Teams tend to add context over time and seldom remove it, so prompts grow heavier month after month. Shorter, clearer instructions and only the documents a task actually needs reduce cost on every call and often improve the quality of the answer. Question to ask: What are we sending to the model on every request that it does not need? Reuse What Repeats When the same instructions or reference material go out with request after request, most major providers offer prompt caching, which stores that repeated content so it is not billed at full price each time. On OpenAI's current flagship models, cached input costs 90-95% less than standard input (OpenAI, n.d.). Anthropic charges a modest premium to store content in the cache but bills each reuse at a tenth of the standard rate or less, so with the standard five-minute cache, the savings begin after a single reuse (Anthropic, n.d.). Shared templates and prompt libraries extend the same idea to people, so teams stop rebuilding the same prompts on their own. Question to ask: Are our most-used workflows set up to take advantage of caching and shared templates? Batch the Work That Can Wait Many tasks do not need an instant answer. Overnight reports, bulk document reviews, and large data-tagging jobs can be submitted in batches and processed in the background at a lower price. Anthropic, for example, offers a 50% discount on both input and output tokens for batch requests and allows that discount to be combined with caching, so the savings on suitable work can be greater still (Anthropic, n.d.). Question to ask: Which of our AI workloads are truly time-sensitive, and which could run overnight at half the cost? Consolidate Overlapping Tools As adoption spreads, different teams often buy similar products to solve the same problem, and individual subscriptions pile up on expense reports outside any central budget. Bringing these under a smaller set of approved tools reduces duplicate fees, strengthens the company's position in vendor negotiations, and lowers security risk, a concern that has grown sharply after this fall's wave of AI-assisted cyberattacks. Question to ask: How many tools are we paying for that do the same job? Manage Tokens Like Any Other Operating Cost A one-time cleanup helps, but the gains last only when leaders can see how AI is being used. Leading companies are beginning to track AI usage the way they track cloud computing or payroll, by team, by use case, and by the business result it supports. Once leaders can see which workflows consume the most tokens and what each one produces, budget conversations become far more productive, spending can follow results, and the teams getting the most value can be given more room to grow. This does not require a large new function. For most mid-sized companies, it starts with three simple habits: naming an owner for AI spending, reviewing each provider's usage reports every month, and asking every team that uses AI to describe what it is getting in return. Those habits are often enough to surface the largest opportunities within a single quarter. Turning Savings Into Advantage The purpose of token optimization is to expand what a company can do with AI. Every dollar recovered from an oversized model, a bloated prompt, or a forgotten subscription can fund the next customer-facing tool, the next automation, or broader training for employees. Over time, an organization that runs its AI efficiently can afford to experiment more often and scale its best ideas faster than competitors spending the same amount with less discipline. As you finalize your 2027 plans, set aside an hour with your finance and technology leaders to work through the five questions above. The answers will likely reveal capacity you are already paying for. For a fuller framework on measuring token use and building it into planning, explore TokenOps: From Token Waste to Competitive Advantage, or reach out to discuss AI strategy and training for your leadership team. References Anthropic. (n.d.). Pricing. Claude Platform Docs. Retrieved November 12, 2026, from https://platform.claude.com/docs/en/about-claude/pricing Apotheker, J., Duranton, S., Lukic, V., de Bellefonds, N., & Schweizer, C. (2026, January 15). As AI investments surge, CEOs take the lead. Boston Consulting Group. https://www.bcg.com/publications/2026/as-ai-investments-surge-ceos-take-the-lead Gartner. (2026, September 18). CEO watch 4Q26: AI investments reveal a dangerous readiness gap. https://www.gartner.com/en/documents/8397281 Larridin. (2026, July 2). How AI tool sprawl is draining your tech budget. https://larridin.com/blog/ai-tool-sprawl-enterprise OpenAI. (n.d.). Pricing. OpenAI API. Retrieved November 12, 2026, from https://developers.openai.com/api/docs/pricing Copyright © 2026 by Severin Sorensen. All rights reserved.

  • Fitting AI to Purpose: Why No Platform Makes You HIPAA-Compliant by Itself

    A business owner in a regulated industry recently asked me a question I expect many leaders are asking. His firm had purchased an enterprise license for Claude, encouraged by others, myself included, who have pointed to the strength of its enterprise data protections. He also hoped that the vendor's HIPAA protections would extend to his firm. Then a peer-group session on cybersecurity left him uneasy about connecting AI to his email and to files containing protected health information. Several colleagues offered what sounded like a sensible fallback: stay with Microsoft Copilot, because it must be safer. That answer deserves examination, because it rests on an assumption that does not survive contact with the documentation. HIPAA compliance cannot be bought with a license. It is the product of fitting a tool to the work it will do, configuring it for that work, and governing how people use it. This holds for Anthropic's Claude, for Microsoft Copilot, and for any platform that follows them. What a Business Associate Agreement actually does When a firm that handles protected health information (PHI) uses an AI vendor, that vendor becomes a subcontractor under HIPAA and needs a Business Associate Agreement (BAA). The BAA assigns legal responsibility for safeguarding the data. It does not configure the firm's account, train its staff, perform its risk analysis, or decide which tasks belong in which tool. Those obligations stay with the firm. Both major vendors offer a BAA, and both draw a boundary around what it covers. For Claude Enterprise, coverage begins only after the organization's Primary Owner enables the HIPAA-ready configuration and accepts Anthropic's BAA. Once that is done, chat, Projects, file work, research, and web search are covered. Cowork, the Claude in Chrome extension, and any data sent to outside services through connectors, including the Microsoft 365 connector, are not. For Microsoft 365 Copilot, Microsoft's BAA comes by default through its Data Protection Addendum, but only for services on its covered list. Microsoft states plainly that web search queries generated by Copilot fall outside the BAA. Since January 2026, Microsoft has also enabled Anthropic's Claude models inside Copilot by default for most US commercial tenants, without publishing a statement that its HIPAA agreement covers that processing. A separate option, "Anthropic models with Data Retention," sits outside Microsoft's terms entirely. As of June 2026, Microsoft's own Purview documentation listed data loss prevention as not yet supported for Copilot Cowork. Neither vendor's agreement covers everything its product can do, and the uncovered features are often the most capable ones. The risk most people are not watching The concern most leaders voice is a breach of the vendor's servers. The risk the research record shows more often is subtler: an AI agent connected to an inbox or shared drive reads everything that arrives, and a crafted email or document can carry hidden instructions telling it to find and send out sensitive data. This is called prompt injection, and it has been demonstrated against both platforms. Examples include EchoLeak (2025) and SearchLeak (2026) in Microsoft 365 Copilot, file-exfiltration research against Copilot Cowork (May 2026), and similar research against Claude's Cowork (January 2026). The vendors have fixed specific flaws. The category of risk remains wherever an agent acts on content it did not author, with the permissions of the person who launched it. A second, quieter risk follows from the same design. On both platforms, the AI can reach whatever the signed-in user can reach. A shared drive with loose permissions becomes, through AI, a very efficient way to surface information people were never meant to see. Sorting the work by data The practical way through is to classify the work by what data it touches before deciding where it runs. I recently did this for an employee benefits consultancy. I took 48 routine tasks across sales, renewals, enrollment, claims advocacy, compliance, HR and payroll, Medicare, and agency operations, and sorted them into four lanes: Full suite. No individual health or sensitive personal data. Marketing, proposals, plan-level comparisons, and template documents fall here. About half the tasks did. Full suite after de-identification. Aggregate or de-identified data that meets HIPAA's Safe Harbor standard. Carrier claims summaries need checking, because "summary health information" may retain five-digit ZIP codes and so still count as PHI. HIPAA-covered features only. Identified PHI, such as large-claim reports, enrollment reconciliation, and claim appeals. Sensitive personal data outside HIPAA belongs here too. FMLA files held by an employer, for instance, are employment records rather than PHI, yet remain protected under ADA confidentiality rules and, in Florida, the Florida Information Protection Act. Decide first. Mixed or unknown sources, such as a whole mailbox, along with systems of record and breach response, call for counsel or a deliberate design decision before any AI touches them. The line between lanes followed the data, not the department. Sales and marketing sat entirely in the first lane, claims advocacy mostly in the third, and renewal work moved between them depending on whether a given file named individuals. Testing the claim that Copilot is safer When I applied the same classification to Microsoft 365 Copilot, 47 of the 48 tasks landed in the same lane. One task favored Copilot: triaging a shared mailbox, because the mail never leaves the firm's Microsoft environment and needs no third-party connector. One favored Claude: comparing a client's prescriptions against available plans, because that work depends on web lookups, and Claude's web search is covered while Copilot's is not. The claim that Copilot is safer, then, is true for some tasks and unsupported as a general rule. The useful question is not which platform is safer, but which platform is safer for which task. Asked that way, the two platforms come out nearly level, and what separates a defensible deployment from a risky one is the discipline the firm brings to it. Six safeguards that apply on any platform Whatever the platform, the same safeguards do most of the work: Confirm the agreement and its scope. Know which services and features your BAA covers, and record the confirmation. Block consumer and personal versions. Free and individual plans, consumer Copilot, and personal accounts sit outside business agreements. Restrict uncovered features for staff who handle PHI. On Claude, that means agent and connector features; on Copilot, web grounding, third-party plugins, and retention-based model options. Check how your platform's role controls apply to administrators, who may not be restricted by them. Review permissions before connecting AI to mail and files. The AI will see what each user can see. Update the risk analysis, training, and incident response plan to cover the platform and the prompt-injection risk. Review vendor terms each quarter. Coverage lists on both platforms changed several times this year. The decision belongs to the firm The leaders I work with do not need a verdict on which vendor is good. They need a way to reason about fit. An AI platform is closer to a set of instruments than a single tool, and each instrument has a proper use, a proper setting, and a proper operator. A firm that classifies its work, matches each task to the right instrument, and governs its use can operate defensibly on either platform. A firm that relies on the brand name, whichever brand it chooses, has not yet done the work that compliance requires. Disclosure: This article was prepared by Severin Sorensen, ePraxis LLC, with research and drafting by AI (Anthropic's Claude, cross-checked with ChatGPT, Gemini, Grok, Perplexity, and Microsoft Copilot) and verified against the vendor and government sources below. AI-assisted research may contain mistakes. This article is informational and is not legal, compliance, or cybersecurity advice; organizations should confirm their obligations with legal counsel. References Anthropic, BAA feature coverage: https://privacy.claude.com/en/articles/8114513 Anthropic, HIPAA-ready Enterprise plans: https://support.claude.com/en/articles/13296973-hipaa-ready-enterprise-plans Microsoft, Enterprise data protection in Copilot: https://learn.microsoft.com/en-us/copilot/microsoft-365/enterprise-data-protection Microsoft, Anthropic models in Microsoft Online Services: https://learn.microsoft.com/en-us/microsoft-365/copilot/connect-to-ai-subprocessor Microsoft, HIPAA/HITECH compliance offering: https://learn.microsoft.com/en-us/compliance/regulatory/offering-hipaa-hitech Microsoft, Purview for Copilot Cowork: https://learn.microsoft.com/en-us/purview/ai-copilot-cowork HHS, Business associates: https://www.hhs.gov/hipaa/for-professionals/privacy/guidance/business-associates/index.html HHS, De-identification guidance: https://www.hhs.gov/hipaa/for-professionals/special-topics/de-identification/index.html HHS, Employers and health information in the workplace: https://www.hhs.gov/hipaa/for-individuals/employers-health-information-workplace/index.html Florida Statutes 501.171: http://www.leg.state.fl.us/statutes/index.cfm?App_mode=Display_Statute&URL=0500-0599/0501/Sections/0501.171.html Copyright © 2026 by Severin Sorensen. All rights reserved.

  • Will We Reach the Singularity by 2026? A Thought-Provoking Journey into AI’s Future

    singularity, a concept popularized by futurist Ray Kurzweil, refers to the point where AI surpasses human intelligence However, these are narrow applications, not the broad Artificial General Intelligence (AGI) that Kurzweil Uncertainties Remain The development of AGI (Artificial General Intelligence) is surrounded by uncertainties

  • Using AI Wisely: What to Share and What to Protect

    In the spring of 2023, engineers at Samsung's semiconductor business pasted confidential source code into ChatGPT while looking for help with routine tasks. Within weeks, Samsung barred staff in one of its largest divisions from using generative AI tools on company devices and networks (Gurman, 2023). The episode became a cautionary tale, yet the behavior behind it has only spread. In Microsoft and LinkedIn's 2024 Work Trend Index, 78% of people who use AI at work said they bring their own AI tools to the job, while only 39% had received AI training from their employer (Microsoft, 2024). Many professionals sense the risk and carry a question they are slightly embarrassed to ask out loud: what is actually safe to type into these tools? The answer becomes far simpler once leaders draw one firm line, which is that business information belongs only in business accounts the organization has approved. With that line in place, AI becomes something teams can use with confidence instead of quiet worry. Why the Account Matters When you type into an AI tool, your words are processed and stored on the provider's servers, and the terms that govern them depend on the kind of account you are using. Business plans are built for organizations. OpenAI states that by default it does not train its models on data from ChatGPT Enterprise, ChatGPT Business, or its API platform, and it offers retention controls to qualifying organizations (OpenAI, n.d.-b). Anthropic's commercial offerings, such as its Team and Enterprise plans, sit outside its consumer terms entirely (Anthropic, 2025). Many business plans also let administrators decide who has access, how long data is kept, and what happens to an account when an employee leaves, and they come with contracts that spell out who owns the data. Personal accounts offer the company none of those protections. Whether conversations are used for training, how long they are kept, and whether reviewers may read them are governed by consumer terms and by settings one employee controls, which can be changed, forgotten, or revisited when terms are updated. Google, for example, notes that human reviewers read a subset of Gemini conversations and asks users not to enter confidential information they would not want a reviewer to see (Google, 2026). The company has no visibility into a personal account, no way to retrieve what is in it, and no contract with the provider, so no configuration makes it appropriate for company work. Any information that belongs to the business, from a client email to a draft strategy memo, should go only into a tool the business has approved. The Cost of Using Whatever Is Closest The risk of ignoring this line is measurable. IBM's 2025 Cost of a Data Breach Report found that one in five organizations had experienced a breach involving shadow AI, meaning AI tools employees use without approval. A breach at an organization with widespread shadow AI costs, on average, $670,000 more than one at an organization with little or none. And, only 37% of organizations have policies to manage AI or detect unapproved use (IBM, 2025). Most of this exposure comes from well-meaning employees reaching for whatever tool is closest at hand. The most useful step a leadership team can take is to provide an approved business plan and ask people to use it for all work, at a cost far below the cleanup after a leak. Five Categories That Deserve Extra Care An approved business account is the right home for work, though not a license to share everything. Some information carries legal or contractual duties that call for caution even inside sanctioned tools. Client and customer details. Names, contact information, account histories, and anything covered by a confidentiality agreement. Confirm that your contracts permit sharing this information with an outside processor, including an approved AI vendor. Unreleased financial information. Earnings before they are announced, forecasts, pricing plans, and anything related to a pending deal. Information like this can move markets and carries legal duties, so keep it within the tools and people cleared to handle it. Personnel matters. Performance reviews, pay, health information, complaints, and investigations. Few things damage trust inside a company faster than learning that an employee's private situation traveled further than it needed to. Passwords and account numbers. Login details, access codes, and bank or card numbers have no place in any AI chat, however secure. Legal matters. Advice from counsel, active disputes, and settlement terms. Sharing them can raise questions only your lawyers can fully weigh, so ask them first. Even within an approved tool, share only what the task requires. Replace names with roles, such as "Client A" or "the regional director," round the numbers, and describe the situation in general terms. A model can help you plan a hard conversation with an underperforming manager without ever learning who that manager is, and the advice will be nearly as good. Put the Line in Writing Every leadership team benefits from a one-page, plain-language AI policy that names the approved tools, states that work information goes only into them, identifies the categories that need extra care, and tells employees whom to ask when they are unsure. Such policies work best when presented as permission, since people readily follow rules that come with a safe, approved way to get their work done. Samsung's engineers were trying to do their jobs well, and the rules about what was off limits arrived only after the leak. A clear rule that business information lives in business accounts, delivered from the top and backed by the right tools, removes the quiet worry that keeps many people from using AI at all. It also protects the trust your clients and employees have placed in you, which is worth far more than any time saved. A Note on Personal Accounts Personal AI accounts remain useful for everyday life, from planning a trip to working through a household budget, and a few points are worth keeping in mind when you use them. Check your training setting. Anthropic lets people on its personal Claude plans choose whether their chats help train future models, and it keeps those chats for up to five years when training is allowed, compared with 30 days when it is not (Anthropic, 2025). ChatGPT's equivalent is a setting called "Improve the model for everyone" (OpenAI, n.d.-a). Use temporary chats for sensitive questions. OpenAI's temporary chats are excluded from training and kept for up to 30 days for safety purposes (OpenAI, n.d.-a), while Gemini holds temporary chats for 72 hours (Google, 2026). Assume a person may read it. Google keeps chats selected for human review for up to three years, even after you delete your activity (Google, 2026), so share personal health, financial, and family details with care. Delete what you no longer need. Anthropic does not use deleted conversations for future training (Anthropic, 2025). References Anthropic. (2025, August 28). Updates to consumer terms and privacy policy. https://www.anthropic.com/news/updates-to-our-consumer-terms Google. (2026, September 24). Gemini Apps privacy hub. Gemini Apps Help. https://support.google.com/gemini/answer/13594961 Gurman, M. (2023, May 2). Samsung bans staff's AI use after spotting ChatGPT data leak. Bloomberg. https://www.bloomberg.com/news/articles/2023-05-02/samsung-bans-chatgpt-and-other-generative-ai-use-by-staff-after-leak IBM. (2025, July 30). IBM report: 13% of organizations reported breaches of AI models or applications, 97% of which reported lacking proper AI access controls [Press release]. https://newsroom.ibm.com/2025-07-30-ibm-report-13-of-organizations-reported-breaches-of-ai-models-or-applications,-97-of-which-reported-lacking-proper-ai-access-controls Microsoft. (2024, May 8). Microsoft and LinkedIn release the 2024 Work Trend Index on the state of AI at work. Microsoft Source. https://news.microsoft.com/source/2024/05/08/microsoft-and-linkedin-release-the-2024-work-trend-index-on-the-state-of-ai-at-work/ OpenAI. (n.d.-a). Data controls in ChatGPT. OpenAI Help Center. Retrieved October 8, 2026, from https://help.openai.com/en/articles/7730893-data-controls-in-chatgpt OpenAI. (n.d.-b). Business data privacy, security, and compliance. Retrieved October 8, 2026, from https://openai.com/business-data/ Copyright © 2026 by Severin Sorensen. All rights reserved.

  • Talk, Don't Type: Why Your Best Prompts Are Spoken

    In a recent article, I made the case that the most useful prompts now run to several hundred lines and read more like a briefing for a new hire than a quick question. The first response I usually hear is a practical one: who has the time to write all of that? The instinct behind the question is understandable. When a busy executive sits down at the keyboard, the prompt tends to shrink to whatever can be typed before the next meeting, and the model is left to guess at everything that was left out. The answer to that problem is surprisingly simple. Stop typing your prompts and start speaking them. Nearly every major AI tool now accepts voice input, and the habit of talking through a request for two or three minutes produces the kind of rich, detailed direction that most people would never take the time to write. For many leaders, it is the easiest way to get the benefits of the long prompt without the burden of writing one. Why Speaking Makes Better Prompts Most people speak about three times faster than they can type on a phone, roughly 150 words per minute compared with 40, so the same few minutes of dictation yields far more material than the same time spent typing (Ruan et al., 2018). Speech recognition has improved considerably since that study, and today’s tools turn spoken words into clean text with remarkably few errors. The second advantage matters more. People explain things differently when they talk. Where a typed prompt tends to state the task and stop, a spoken one wanders in useful directions, picking up the background, the history, the concerns, and the small details that an experienced leader carries in their head but rarely writes down. Ask an executive to type a request for a board update and you may get one line. Ask the same person to talk it through and you will hear who sits on the board, which member always presses on cash flow, what went poorly last quarter, and what tone the CEO wants to strike. That second version is exactly the context a model needs to do excellent work. The major AI developers say the same thing in their own guidance. Clear, detailed direction that explains the audience, the purpose, and what a good result looks like produces far better output than a short instruction (Anthropic, n.d.; OpenAI, n.d.). Speaking is simply the fastest way most people have of giving that detail. There is also a third, quieter benefit: many leaders think best out loud. Talking a problem through to a model often clarifies the request in the speaker's own mind, and the model can then act on thinking that might never have made it onto the page. How to Give a Spoken Briefing Start by downloading the mobile app for whichever generative AI tool you already use, whether that is ChatGPT, Claude, Gemini, or something else. These tools offer their own microphone buttons on desktop and mobile alike, but your phone’s built-in dictation, which you can reach through the microphone icon on the keyboard, tends to produce the most reliable transcription and is worth using instead. Dictate on your phone, then pick up the conversation on your desktop when it is time to review the work. Beyond adjusting to moving between your phone and your desktop, the hardest part is breaking the habit of short requests. A little structure helps, and the easiest approach is to talk to the model the way you would brief a capable new member of your team. A good spoken briefing usually covers five things: Who is involved. Tell the model what role to play, describe the audience and what they care about most, and fill in the background on your organization that a capable outsider would lack. Why it matters. Explain the task and the decision or outcome it supports, since a model that understands the purpose makes better judgment calls than one following bare instructions. What good looks like. Describe the voice, length, and format you want, and walk through a past example you were pleased with. What to avoid, and why. Name the mistakes, sensitive topics, and unusual situations that have caused problems before, and explain the reasoning behind each so the model can apply it to cases you did not anticipate. What you are unsure about. Acknowledge where your own knowledge runs out, ask the model to approach the work as a seasoned expert, and invite it to raise the risks, assumptions, and questions you may have overlooked. Then close with one simple instruction: "Before you start, ask me any questions you need answered." This gives the model a chance to fill the gaps that even a thorough spoken briefing leaves behind. Do not worry about speaking in polished sentences. Today's models are very good at finding the meaning in natural, rambling speech, including the false starts and the "actually, let me back up" moments. If a request is complex, you can even ask the model to first turn your spoken notes into a clean, organized brief, review that brief, and then have it begin the work. Leaders who find a briefing worth keeping can save it and reuse it, which over time builds the kind of prompt library I described in the earlier article. A Few Sensible Cautions Speaking a prompt calls for the same judgment as typing one, along with a few added habits. Be mindful of where you are. A detailed briefing about a client or a personnel matter should not be delivered in an airport lounge or a crowded coffee shop, where the people nearby can hear every word. Take the same care with what you share with the tool itself, since a spoken prompt reaches the model just as a typed one does. It also pays to glance at the transcript before you send it, because names, figures, and industry terms are where dictation most often slips. Finally, remember that more words are only valuable when they carry useful information. The aim is to pass along the context that lives in your head and would otherwise never reach the model, which usually takes a few focused minutes rather than a long monologue. In TokenOps terms, a well-structured three-minute briefing that gets the work right the first time costs far less than a vague one-line request followed by four rounds of correction. The Leadership Habit Behind It Speaking is the most natural way most of us explain what we want and why, and dictating a prompt turns what would be a writing chore into a conversation. Try it once this week. Before your next call or on a walk between meetings, open your AI tool of choice, tap the microphone on your keyboard, and brief it the way you would brief a trusted colleague. The difference in what comes back is likely to change how you work with AI from then on. References Anthropic. (n.d.). Prompting best practices. Claude Platform Docs. https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/claude-prompting-best-practices OpenAI. (n.d.). Best practices for prompt engineering with the OpenAI API. OpenAI Help Center. https://help.openai.com/en/articles/6654000-best-practices-for-prompt-engineering-with-the-openai-api Ruan, S., Wobbrock, J. O., Liou, K., Ng, A., & Landay, J. A. (2018). Comparing speech and keyboard text entry for short messages in two languages on touchscreen phones. Proceedings of the ACM on Interactive, Mobile, Wearable and Ubiquitous Technologies, 1(4), Article 159. https://doi.org/10.1145/3161187 Copyright © 2026 by Severin Sorensen. All rights reserved.

  • Research Review: Evaluating the Effectiveness of AI vs. Human Coaching in Achieving Goals

    As the realm of artificial intelligence (AI) continues to expand into various domains, the question of experience the benefits of coaching through AI, their exposure could foster demand for more advanced and intelligent “Systemic eclectic”: Coaches that can “exhibit a sensitive, intelligent approach to the client situation While AI coaching's mechanistic execution of goal theory compensates for its lack of nuanced intelligence Comparing artificial intelligence and human coaching goal attainment efficacy.

  • Why the 300-Line Prompt Has Become the New Normal

    Only a few years ago, working with an AI model meant typing a sentence or two and hoping for something useful in return. Early adopters traded tips about magic phrases and clever openings, and the prevailing assumption was that a good prompt was a short one. That assumption has quietly collapsed. The most effective teams now routinely write prompts that run to several hundred lines, and those prompts read far less like questions and far more like operating manuals. Leaders who still picture prompting as a clever one-liner are underestimating both what the technology can do and what it asks of the people directing it. What Changed Three changes came together to make long prompts practical and, increasingly, necessary. The first is capability. Today's models can follow detailed instructions far more reliably than earlier versions could, which means the limiting factor in output quality has moved from the model to the clarity of the direction it receives. The second is capacity. Models can now take in enough text at once to hold extensive instructions, reference material, and examples without crowding out the work itself. The third, and perhaps most important, is the nature of the work. Organizations have moved from asking AI for one-off answers to building it into repeatable, everyday work such as drafting proposals, sorting customer inquiries, summarizing research, and producing first drafts of recurring reports. When a prompt will run hundreds or thousands of times, the math behind writing it changes entirely. Spending an afternoon refining instructions that will shape every future output is a sound investment, in the same way that a thorough onboarding program pays for itself many times over across a new employee's tenure. That onboarding comparison is the most useful way to understand what a long prompt actually is. Imagine hiring a brilliant analyst who arrives on the first day knowing a great deal about the world and nothing at all about your company, your clients, your standards, or the reasons behind your decisions. No sensible manager would hand that person a sticky note reading "write the quarterly report" and expect excellent results. The manager would explain who reads the report, what they care about, what a strong version has looked like in the past, which mistakes have caused problems before, and how to handle the situations that inevitably fall outside the norm. A well-constructed 300-line prompt is that briefing, written down once and delivered consistently every time. What Fills Those Lines The substance of a mature prompt tends to follow a recognizable pattern. It establishes the role the model should play and the audience it is serving, then supplies the organizational context a capable outsider would lack. It describes the task along with its purpose, because a model that understands why a task matters makes better judgment calls than one following bare commands. It sets standards for voice, format, and quality, offers worked examples of excellent output, and addresses the unusual situations where unclear direction tends to cause mistakes. The best prompts also explain the reasoning behind their rules, since models apply the thinking behind a rule far more reliably than a bare list of things to avoid. A rule that says to avoid jargon becomes much more effective when it explains that the readers are time-pressed executives outside the field. Increasingly, the most sophisticated prompts also include an admission of their author's limits, acknowledging that the person writing them does not know what they do not know. These prompts ask the model to act as a seasoned expert in the domain, to look for what the author cannot see, and to surface the risks, assumptions, and overlooked questions that someone with deeper experience would raise. That single instruction invites the model to help improve the brief as well as carry it out. The Token Question Readers familiar with my work on TokenOps may reasonably ask whether all of this contradicts the discipline of token efficiency. It does not, because token waste should be judged by the results a prompt produces. A short prompt that produces output requiring three rounds of correction consumes more tokens, and considerably more human attention, than a thorough prompt that delivers usable work on the first attempt. Many AI platforms also store repeated instructions so they can be reused at a much lower cost. The real sources of waste in long prompts lie elsewhere, in repeated instructions that say the same thing three ways, contradictory rules accumulated over months of patching, and outdated guidance that no longer reflects how the organization works. Token discipline in this new environment means keeping every line purposeful, and it has very little to do with keeping the total prompt word count small. Length Comes From Detail Prompt length comes naturally from being specific, and a prompt padded with vague encouragement or words in all capital letters will underperform a shorter one written with precision. The failure patterns are consistent: rules pile up without anyone reconciling them, examples are so narrow that the model imitates them too literally, and instructions multiply without the reasoning that would help the model apply them sensibly to situations nobody anticipated. The remedy is to treat important prompts as living documents with clear ownership. Someone should be accountable for each one, changes should be tracked, and revisions should be tested against a realistic set of sample requests before they go into production. Organizations already apply this kind of rigor to policies, contracts, and code, and prompts that shape customer communication or strategic analysis deserve the same care. What Leaders Should Do Now Prompt writing should be recognized as a professional skill that combines subject knowledge, clear writing, and organized thinking, and it deserves investment and development accordingly. High-value prompts should be treated as organizational assets, kept in a shared library where good practice can spread across teams. Their performance should be measured by the quality and consistency of what they produce, with the same seriousness applied to any other operational process. The organizations pulling ahead in this period share a quieter advantage: they have learned to put what they know into words. The 300-line prompt is ultimately an exercise in writing down the know-how that usually goes unspoken, capturing the judgment that lives in the heads of experienced people and putting it into a form that can be shared, tested, and improved. That discipline makes AI more useful, and it tends to make the organization itself sharper in the process. Copyright © 2026 by Severin Sorensen. All rights reserved.

  • Picking the Right AI Model for the Job

    Most professionals choose one AI model, set it as their default, and send every request through it, whether the task is rewording a two-line email or stress-testing a five-year growth plan. The habit feels efficient, yet it has become one of the quietest sources of waste in how organizations use AI. Sending routine work to a top-tier reasoning model is a bit like asking your CFO to reconcile petty cash, and sending strategic work to a lightweight model is like handing the board deck to a summer intern. Each mismatch carries a cost, the first in dollars and time and the second in the quality of the thinking you receive. How Model Tiers Work Every major AI provider now organizes its lineup in a similar way, pairing fast, inexpensive models for high-volume everyday work with slower, deeper reasoning models for problems that reward careful thought. Anthropic offers Claude Haiku at the fast end, Sonnet in the middle, and Opus and its newest premium model, Fable, at the top. OpenAI splits GPT-6 into Luna, Sol, and Astra. Google pairs Gemini Flash and Flash-Lite with Pro and Deep Think. xAI follows a comparable pattern with its Grok models. Between those two poles sits a capable middle tier that deserves more attention than it usually receives. Models such as Claude Sonnet 5.5 and GPT-6.1 Sol now deliver quality close to the frontier at a fraction of the price, which makes them a sensible default for most professional work, from drafting a proposal to analyzing a quarterly report. A practical way to choose among the tiers is to ask what it would cost you if the answer were slightly wrong. When an error would be trivial and easy to spot, as with a meeting recap or a rewritten subject line, a fast model is the right call and will return its work in seconds. When an error would matter but would likely surface on review, as with a client proposal or a market summary, the middle tier offers the best balance of quality and price. When an error would be expensive or difficult to detect, as with a pricing strategy, a contract review, or a narrative for the board, the reasoning model earns its higher price many times over. Which Model for Which Job The table below reflects the major lineups as of October 2026. Names change almost monthly, so treat the tier as the lasting lesson and the specific version as a snapshot. Task Good fits Why this tier Quick drafts, email replies, rewrites, meeting recaps GPT-6 Luna; Gemini 3.8 Flash; Claude Sonnet 5.5 (Haiku 4.5 as an interim option until Haiku 5.5 ships) Fast and inexpensive, with quality that is more than sufficient for low-stakes work Everyday professional work (the new default) Claude Sonnet 5.5; GPT-6.1 Sol; Gemini 3.8 Flash Near-frontier quality at a mid-tier price, making this the right starting point for most knowledge work Executive communications and long-form writing Claude Sonnet 5.5 or GPT-6.1 Sol; Claude Opus 5.5 when the piece is sensitive or consequential Stronger command of tone, structure, and nuance across longer pieces Strategy, complex analysis, high-stakes decisions Claude Opus 5.5 (Fable 5.1 when results still fall short); GPT-6 Astra or GPT-6.1 Sol at high effort; Gemini 3.1 Pro or 3.1 Deep Think Deep reasoning that weighs tradeoffs and catches flawed assumptions, best paired with a second-model critique and human review Coding and AI that completes multi-step tasks on its own Claude Opus 5.5 or GPT-6.1 Sol; Claude Fable 5.1 or GPT-6 Astra for the hardest work; Gemini 3.8 Flash for cost-sensitive tasks Built to work through long, involved tasks one step at a time, with the strongest reported coding results at the top tier Long videos, audio, large documents, and visual content Gemini 3.8 Flash, then a stronger reasoning model when conclusions matter Native video and audio input at a low price; document length alone is no longer a differentiator now that Claude also handles very large files Real-time news and social sentiment Any strong model with search enabled; Grok 4.7 with X Search when X is the primary source Current information comes from search tools rather than the model itself, so findings should be corroborated against established news sources High-volume, cost-sensitive processing GPT-6 Luna; Gemini 3.5 Flash-Lite Very low usage costs for sorting, labeling, and summarizing large batches, with uncertain cases routed to a stronger model Customer-facing or regulated workflows An economy or default-tier model behind retrieval, guardrails, and audit logs Operational design matters more than model choice here, with a human handoff for disputes and risk signals A Simple Routing Habit You do not need a technical team to put this into practice, since most AI apps include a menu for choosing the model. Start everyday tasks on the middle tier, move routine requests such as recaps and quick replies to the fast tier, and rerun the prompt on the reasoning model when an answer feels shallow or is headed to a client or your board. On larger projects, let the faster tiers handle groundwork such as pulling figures and drafting slide narration, and save the reasoning model for the questions leadership will press on. For recurring work, write down which model handles each step and revisit those choices as vendors release new versions. Workflows that touch customers or regulated data call for more than a model choice, and they are worth designing with your technology and compliance teams so that safeguards and human review are built in from the start. Within a few weeks, the decision will feel as natural as choosing between a quick message and a full meeting. Model Choice as a Leadership Discipline Matching each task to the right model is the simplest place to begin, and it requires nothing more than a moment of judgment before you press enter. Made consistently, that small decision lowers costs and sharpens the thinking behind the decisions that matter most. Leaders who make this choice visibly, and talk it through with their teams, also send a useful signal that AI is a set of tools to be matched to the work, each with its own cost and its own strengths. Copyright © 2026 by Severin Sorensen. All rights reserved.

  • How ChatGPT Can Save You 15+ Hours Each Week

    In recent years, artificial intelligence (AI) has emerged as a game-changer in this quest for efficiency

  • The Feedback Loop No One's Coaching For: Giving Performance Reviews to an AI Employee

    Most executives can now name the AI agent handling their expense reports, drafting their first-pass contracts, or triaging their inbox. Fewer can say what a “good quarter” looks like for that agent, or what happens when it quietly underperforms. While executive teams have gotten comfortable deploying AI colleagues, they have not yet gotten comfortable evaluating them. The scale of the shift explains the urgency. Gartner projects that by the end of 2026, 40% of enterprise applications will embed task-specific AI agents, up from under 5 percent in 2025 (Gartner, 2025). Field research from MIT’s Initiative on the Digital Economy is already documenting how teams that work alongside AI agents perform differently on real tasks, not hypothetical ones (Ju & Aral, 2025). Agents have moved from pilot projects to production, but what has not moved at the same pace is the discipline of reviewing their work the way a manager reviews a direct report’s. Providing AI Feedback Most leaders were trained, at some point, on a model for delivering feedback to a human being. The Situation-Behavior-Impact model, developed by the Center for Creative Leadership, is among the most widely used: name the situation, describe the specific behavior, and state its impact, keeping interpretation and personality out of the conversation entirely (Center for Creative Leadership, 2025). The model works because it forces precision. For example, “You were unreliable this month” invites an argument. Whereas, “In the March renewal cycle, you flagged four contracts as low-risk that later required legal review, which cost the team roughly a week of rework” invites a conversation about what to change. That same discipline transfers cleanly to an AI agent, with one addition. A human behavior usually has intent behind it worth exploring, which is why CCL’s extended model adds a fourth step: asking about intent to turn feedback into dialogue (Center for Creative Leadership, 2025). An AI agent has no intent in that sense, but it has something that plays a similar role: the configuration that produced the behavior. Call it Situation, Behavior, Impact, and Root Cause. Situation names the specific task and context, not “the chatbot” in general but the exact workflow, prompt chain, or trigger that ran. Behavior describes what the agent actually did, in terms as observable as a transcript or an output log allows, resisting the temptation to say the agent “decided” or “chose” when what happened was closer to “produced.” Impact states the downstream effect in the same terms an executive would use for a human report: hours saved or lost, revenue affected, risk introduced, trust gained or spent with a client or colleague. Root Cause is where the review does its real work, tracing the behavior back to its source, whether that is the underlying prompt, the tools the agent had access to, the data it was trained or grounded on, or the absence of a guardrail that should have caught the error before it reached a client. Root Cause is also the hardest step, and the one most organizations have not learned to do well, because it cannot be answered the way a human review answers it. With a person, you can simply ask why. An AI agent cannot answer that question in kind, so the root cause has to be reconstructed rather than asked for, which means actually having the underlying prompt, the tool logs, the data the agent was grounded on, and a clear sense of what guardrail should have caught the error before it reached a client. Most organizations running agents in production do not yet keep that record in a form anyone can review after the fact, which means the step of the model built to turn a finding into a fix is the one most reviews quietly skip. That gap compounds in a multi-agent workflow, where one agent's output becomes another agent's input before a human ever sees the full chain. An executive should be able to answer a simple question walking into any review: if this chain produces a bad outcome, whose name goes on the corrective action. Research on agentic governance suggests most organizations cannot answer that question with confidence today (Cloud Security Alliance, 2026), the same ambiguity a coach would flag immediately in a human org chart, but one that AI's speed and opacity make far easier to leave unresolved. The Quarterly Review The record-keeping problem points toward a solution: review more often, in smaller batches, while the trail is still traceable. An annual AI review, the cadence most organizations default to for talent, asks a reviewer to reconstruct root cause across a year of logs and prompt changes, which is close to impossible and explains why so many of these reviews never happen at all. A quarterly AI performance review, run with the same seriousness as a talent review and folded into the same calendar, keeps the sample small enough that a root cause is still findable and the config that produced it likely has not changed twice since. For each agent operating with meaningful autonomy, the executive team or a designated owner should walk through the Situation-Behavior-Impact-Root Cause structure against a sample of its highest-stakes work from that quarter, name what changed since the last cycle, and make one of three calls: continue as configured, retrain or reconfigure with a specific root cause attached, or retire the agent from that task entirely. The point is to bring the same rigor that prevents human underperformance from going unaddressed for a year to a category of colleague that can now do a year's worth of damage in a week. The Main Takeaway The organizations that get this right will be the ones whose leaders treat evaluating an AI agent’s work as seriously as they treat evaluating a person’s, because the coaching skill underneath both is the same: specific, evidence-based feedback that someone is accountable for acting on. That skill was never really about humans; it was about closing the loop between what happened and what happens next, and right now, for most AI agents in most organizations, that loop is still open. References Center for Creative Leadership. (2025). SBI feedback model & talent development conversations. https://www.ccl.org/articles/leading-effectively-articles/sbi-feedback-model-a-quick-win-to-improve-talent-conversations-development/ Cloud Security Alliance. (2026). NIST AI Risk Management Framework: Agentic profile. https://labs.cloudsecurityalliance.org/agentic/agentic-nist-ai-rmf-profile-v1/ Gartner Predicts 40% of Enterprise Apps Will Feature Task-Specific AI Agents by 2026, Up from Less Than 5% in 2025. (2025). Gartner. https://www.gartner.com/en/newsroom/press-releases/2025-08-26-gartner-predicts-40-percent-of-enterprise-apps-will-feature-task-specific-ai-agents-by-2026-up-from-less-than-5-percent-in-2025 Ju, H., & Aral, S. (2025). Collaborating with AI agents: Field experiments on teamwork, productivity, and performance. MIT Initiative on the Digital Economy. https://ide.mit.edu/ Copyright © 2026 by Severin Sorensen. All rights reserved.

bottom of page