Confluence for 11.16.25
OpenAI releases GPT-5.1. Anthropic's use case library. Developing Uniquely Human Skills. AI progress depends on more than tech.
Welcome to Confluence. Here’s what has our attention this week at the intersection of generative AI, leadership, and corporate communication:
OpenAI Releases GPT-5.1
Anthropic’s Use Case Library
Developing Uniquely Human Skills
AI Progress Depends on More than Tech
OpenAI Releases GPT-5.1
Early testing suggests meaningful improvements in writing quality.
If the end of 2025 is anything like the end of 2024, we can expect a flurry of releases from the leading generative AI companies in the next six weeks. OpenAI started things off last Wednesday with the release of GPT-5.1, which is rolling out to all ChatGPT users in “Instant” and “Thinking” variations. According to the official OpenAI release, GPT-5.1 “improves meaningfully on both intelligence and communication style” and shows improvements in instruction following.
It’s still early, but initial third-party testing suggests that GPT-5.1 does represent a step up in conversational style and writing capabilities. EQ-Bench, a set of benchmarks focused on non-technical LLM capabilities, shows GPT-5.1 (still shown in the benchmark ranking by its codename, polaris-alpha) now ranked #1 for creative writing and #2 (behind Claude Sonnet 4.5) for long-form writing.
Given the late-week release, we have not yet had time to fully put 5.1 through its paces. But we did conduct a quick, rough test of GPT-5.1’s writing capabilities compared to GPT-5. We gave each model three prompts, moving from simple to more complex:
Please give me a one-paragraph summary on media richness theory and why it matters for executives.
This past week OpenAI released GPT-5.1. Write a three-paragraph summary of the release. Use the first two to talk about GPT-5.1’s technical abilities, and use the final paragraph to comment on the potential implications of this release.
Write an essay on how the concept of “wins above replacement” in baseball could be applied to knowledge work as generative AI becomes more capable and widespread.
We then asked Claude Sonnet 4.5 to blindly rank the output of each, with a simple prompt of “Which of these do you think is better-written and why?” Claude ranked 5.1’s output as superior in all three cases, and we agreed. The differences were not dramatic, but they were noticeable.
GPT-5.1’s output for prompt #3—“Write an essay on how the concept of ‘wins above replacement’ in baseball could be applied to knowledge work as generative AI becomes more capable and widespread”—was particularly interesting. We’ve included the full initial output in a footnote1. We decided to take it further with a second prompt:
Revise this so it’s entirely in prose paragraphs. Use bulleted lists sparingly and only when necessary. This should have a title and you can use a few sub-headings where necessary and additive.
We’ve included that revised output in its unedited entirety below both as a demonstration of GPT-5.1’s smarts and writing ability and for the value of the insights.
Wins Above Replacement in the Age of AI
In baseball, wins above replacement (WAR) offers a clarifying way to think about value. A player isn’t evaluated against perfection but against a “replacement-level” alternative—a competent, generally available player who can be obtained at low cost. WAR asks a simple question: How many more wins does this player contribute compared with someone who is easy to replace? As generative AI becomes more capable and widespread across knowledge work, this framing becomes newly relevant. The baseline for competent performance is rising, and as it rises, the definition of differentiated value changes with it.
The Replacement Level Is About to Rise
For most of recent history, the “replacement-level” knowledge worker was defined by human constraints: average skills, average judgment, and average availability. Even tasks that required little creativity or judgment still required human time and attention. Generative AI alters this equation. When a system can produce reasonable drafts instantly, synthesize research, organize information, and generate plausible options across domains, the replacement baseline shifts upward. The standard is no longer the average junior analyst or communicator—it is now the combination of an average human plus a capable AI system. In this context, contributing real value means outperforming that new baseline, not merely completing tasks competently.
As Baselines Rise, Value Moves Upstream
When performance improves across an entire sport, as happened in baseball with analytics and better training, the relative value of certain skills doesn’t disappear—it moves. The same thing is happening in knowledge work. If AI can reliably handle first drafts, background summaries, or structured problem exploration, the locus of human value shifts toward upstream capabilities: asking sharper questions, recognizing organizational dynamics, discerning which trade-offs matter, and framing decisions so leaders can act with confidence. These are subtler, context-rich skills that resemble pitch sequencing or defensive positioning in baseball—skills that rarely get measured directly but that change outcomes profoundly.
In this environment, humans stand out not by competing with AI on task execution, but by excelling at the judgment, nuance, and relational intelligence that determine which tasks matter and how they should be approached.
Context Turns WAR into a Human-Centric Metric
Baseball’s WAR is designed to be context-free. In organizational life, context governs everything. Two analysts with identical AI-augmented technical output may produce very different “wins” depending on their relationships, credibility, understanding of the culture, and ability to anticipate what stakeholders actually need. In many cases, the value lies not in the artifact produced but in the conversations shaped, the alignment created, or the confidence instilled.
This means that WAR in knowledge work is less about what you produce and more about how your presence improves the performance of others. Once AI handles the baseline execution, the differentiators become influence, coordination, problem framing, and leadership—the inherently human capacities that turn information into action.
AI Raises the Floor Faster Than It Raises the Ceiling
One of the overlooked implications of AI’s advance is the compression of the lower end of the performance distribution. Workers who once delivered uneven or inconsistent output can now deliver competent, AI-assisted work with relatively little effort. Teams become more reliable, the minimum standard rises, and variance shrinks.
But raising the ceiling—elite-level insight, nuanced judgment, or relationship-driven influence—remains far more difficult. These qualities develop through experience, reflection, and tacit knowledge. As a result, the WAR analogy reinforces the importance of uniquely human strengths. The most valuable contributors will be those who can orchestrate teams of people and models, discern which AI outputs are genuinely useful, and integrate insight with context, stakes, and interpersonal realities.
A Managerial Lens on WAR
For leaders, the WAR framing becomes a practical diagnostic. It encourages a clear-eyed look at how roles are structured and where human talent is currently being spent. Many organizations still assign people to tasks that AI can now perform at replacement level. Using WAR as a lens invites questions such as: Where do we truly need human judgment? Where is human attention being diluted by routine tasks that no longer represent differentiated value? Where can AI elevate the contribution of each team member rather than simply replacing a part of their workload?
The aim is not to depersonalize work but to design roles and workflows that allow people to apply their highest-value skills. When organizations do this effectively, they expand the number of “wins” each person can contribute.
Managing One’s Own WAR Curve
For individuals, the concept becomes personal and forward-looking. The central question becomes: In what areas do I contribute value that cannot be easily replaced—or where my ability to use AI allows me to contribute significantly more? This encourages a deliberate investment in capabilities that remain uniquely human or become more valuable in an AI-rich environment: discerning judgment, narrative skill, relationship-building, and the ability to diagnose ambiguous situations.
It also places a premium on becoming an orchestrator of AI systems rather than a competitor to them. Workers who can frame problems well, interpret outputs, and integrate AI effectively will deliver more wins for their teams and organizations than those who simply produce more volume.
A Final Observation: WAR Measures Contribution, Not Activity
The most important lesson from the analogy may be the distinction between effort and contribution. WAR doesn’t care how many swings a player takes; it measures how much they help the team win. As AI takes on more activity—the drafting, synthesizing, sorting, and generating—the core human contribution becomes clearer. Insight, clarity, persuasion, and trust rise in relative importance. The value lies in shaping decisions, diagnosing context, and enabling others to perform at their best.
In this way, the analogy doesn’t diminish the role of people in an AI-first environment. It sharpens it. A rising replacement baseline forces individuals and organizations to define where real human value lives. Those who understand their own WAR—where they add wins beyond what a readily available AI can deliver—will be positioned to thrive as the nature of knowledge work continues to change.
We’ll continue testing GPT-5.1 and will share further insights in future editions of Confluence. We’d encourage any of our readers who use ChatGPT to put it through its paces as well, as these incremental improvements can make a significant difference for real-world utility. We expect this will be the first of many significant releases in the coming weeks, so stay tuned.
Anthropic’s Use Case Library
Another way to open your aperture to what’s possible.
One of the best ways to learn what generative AI can do is to watch others use it in new or unexpected ways. Even when these use cases don’t translate directly to your needs, they open your eyes to what’s possible. Take the idea of asking your LLM of choice to interview you about your goals before writing a memo. You may not need to write a memo tomorrow, but seeing this use case in action gives you ideas for when it might be helpful to ask the LLM to interview you before producing a different type of output (say, a travel itinerary or a prompt for a Project or Custom GPT).
Anthropic recently launched its own use case library, and it’s the best version of this idea we’ve seen. They’ve purposefully curated the library to showcase a broad range of capabilities with Claude. The way you interact with the use cases is smart, too. You can see the specific prompts Anthropic uses and with a single click, open a new Claude tab with the prompt pre-populated in the window. You’ll want to tweak the prompts before running them, but the results can impress. For instance, we took the initial prompt for the “create a custom webpage” use case, revised it to be about Philadelphia, and with a single “make it better” follow-up prompt we were able to create this:
There are plenty of other use cases in this library from which to learn. Pick a few that feel close to a real opportunity you have and experiment. There is no exhaustive catalog for all the generative AI use cases that exist, and each person should be working to hone their skills and fluency with generative AI to the point where they can learn along the way. When we see how others use the technology, we should be able to 1) understand and break down why such a use case works and 2) identify how we might apply the learnings in our own day-to-day. Not every experiment is going to pan out, but each experiment will teach you something about the strengths and limitations of this technology.
Developing Uniquely Human Skills
Using AI can help knowledge workers prove their human value.
We spent some time with a client this week who is thinking carefully about what uniquely human skills their teams will require to add clear value in the age of AI. We identified many of the same capabilities the GPT-5.1 piece on WAR describes above: employees will need to have robust business acumen, know how to ask smart questions, consider second- and third-order consequences, and build strong relationships. The client is working out how to develop these capabilities at scale.
As an example, the client referenced high-performing risk analysts who, largely because of the highly strategic nature of their work, have built a regular practice around many of those key skills. Ideally, the organization could deploy this kind of in-depth analysis across every region of the business. They wondered if the solution might simply be to build AI agents to do that analysis for each regional office, or if this should be one of those skills they preserve as a key human skill.
The conversation struck us because it’s a good example of how discussions like this can create a false divide: uniquely human skills on one side, and tasks that AI can clearly augment or automate on the other.
But this framing misses something important. Yes, we all need to be as clear-eyed as possible about what human skills will matter most in the future. But we needn’t let that “unique humanness” prevent us from thinking equally as seriously about how AI can help employees gain, practice, and create meaningful habits around those skills. Rather than build an agent that does the work of risk analysis, we could build one that helps employees develop related habits—like a tool designed to help users think through potential consequences of strategic decisions or work through crisis scenarios. Anthropic’s use case library, which we write about above, already includes a few tools—like this one for debate practice—designed to develop, rather than replace, skills like this.
This kind of use case isn’t new, but as the pace of work speeds up and opportunities for early-career employees to gain essential experience (might) shrink, it’s a useful reminder that AI will be an important tool for helping employees excel at higher-order human skills. As we all adjust, in ChatGPT’s words above, to the shifting “locus of human value,” AI will, ironically, be a key component in helping employees develop the skills they’ll need to prove their value beyond it.
AI Progress Depends on More than Tech
A recent Wall Street Journal article offers some important perspective.
By some estimates, the capabilities of generative AI models are doubling about every seven months. Every organization in which we work is at least talking about their use of generative AI, and many are making significant investments in the technology. And the world’s leading technology firms are building AI-supporting infrastructure at a level of spending that rivals the Manhattan Project. Which is why we read this recent Wall Street Journal piece with such interest: When AI Hype Meets AI Reality: A Reckoning in 6 Charts.
The overall point of the article is that independent of the technical advancements of generative AI, its evolution and application depend on physical constraints that have much different time and cost scales. For some of you the article may be behind a paywall, so we’ll summarize the six charts in a few bullets:
The data center gold rush is colliding with reality. Firms have planned the construction of a massive number of AI data centers, but many are stalling or halted because land speculators jumped in without understanding the complexity of these projects. Securing power connections, fiber access, and actual tenants is proving far harder than acquiring land. The gap between “planned” and “under construction” reveals how physical infrastructure challenges are overwhelming even unlimited capital.
AI spending is fundamentally reshaping tech company economics. The four major tech companies—Amazon, Microsoft, Alphabet, and Meta—are collectively spending over $100 billion quarterly on AI infrastructure in 2025, approaching 40% of their total revenue. This is a level of core operational spending that consumes nearly half of what these companies earn, creating unprecedented pressure to generate returns.
Manufacturing and energy production bottlenecks matter more than money. Goldman Sachs projects global data center capacity will grow from 71 gigawatts today to 109 gigawatts by end of 2027. But this is conservative compared to other estimates. Why? Transformer shortages and sources of energy production. The step-down transformers that connect buildings to power grids are in such short supply, and the demand for equipment like natural gas turbines so high, that GE Vernova has orders booked through 2028. The constraint is that one simply cannot manufacture this equipment any faster, making this a 10-15-year infrastructure problem.
The physical footprint demands are extraordinary. Data centers will occupy 645 million square feet by 2027—several multiples greater than every Costco warehouse in existence combined. This spatial requirement compounds every other constraint: you need not just the square footage, but the right locations with power access, fiber connectivity, and cooling infrastructure. Each dependency creates cascading delays that slow even the most capitalized projects.
The revenue math reveals the scale of the collective bet. JPMorgan analysts calculated that $5 trillion in global AI infrastructure investment through 2030 requires $650 billion in new annual revenue—indefinitely—to generate a reasonable 10% return. To put that in perspective: that’s equivalent to every iPhone owner worldwide paying an extra $35 per month for AI products and services, forever. The consumer and enterprise market for AI at this scale doesn’t yet exist.
The pace of development in generative AI is astonishing, but it still exists within a larger physical context, and tracking that context is important to making informed assumptions about where the technology goes and what happens next. As the Journal notes:
AI is already dramatically transforming our lives and businesses. But the real world limits how quickly companies can scale up these next-level supercomputers, and it’s unclear who will pay for all the resulting services. It would be wise to moderate our expectations—or at least adjust the timetable—for the AI revolution.
We’ll leave you with something cool: Anthropic shared lessons learned from Project Fetch—its internal competition to create a robot dog who can, well, fetch (among other tasks).
AI Disclosure: We used generative AI in creating imagery for this post. We also used it selectively as a creator and summarizer of content and as an editor and proofreader.Wins Above Replacement for the Age of AI
In baseball, wins above replacement (WAR) offers a simple, clarifying idea: a player’s value is measured not against perfection, but against a “replacement-level” player—someone generally available, competent, and inexpensive. WAR doesn’t ask whether a player is great in the abstract; it asks, How many more wins does this person contribute compared with the readily available baseline?
As generative AI becomes more capable and more pervasive across knowledge work, the WAR framing becomes newly relevant. The major shift underway is that the “replacement-level” baseline for many tasks is rising dramatically. And once the baseline rises, the definition of differentiated value—what truly adds “wins”—changes with it.
Below are several ways the analogy helps make sense of the transition.
1. The Replacement Level Is About to Jump
For most of the past century, the “replacement-level” knowledge worker was anchored in human constraints: average skills, average judgment, average availability, average consistency. Even the lowest-value tasks required manual time and attention.
Generative AI resets this baseline. A system that never tires, executes instantly, works across domains, and learns rapidly is becoming the minimum standard for a wide range of activities: drafting a memo, synthesizing background research, brainstorming options, organizing data, or generating a range of plausible solutions.
This doesn’t mean human workers are obsolete. It means that for a growing set of tasks, the replacement-level performance is no longer a junior analyst or an average communicator—it’s a capable model that delivers high-competence, low-cost outputs on demand. The WAR calculation shifts accordingly: contributing value requires beating the combination of human-plus-AI that is now generally available.
2. When Baselines Rise, Value Moves Upstream
In sports, when every player improves—through analytics, nutrition, training—the relative value of skills doesn’t disappear; it shifts. The same is happening in knowledge work.
If AI can competently handle the “scouting report” level work—summaries, pattern recognition, categorization, first-draft generation—then the value of a human advisor moves to higher-order layers:
Asking sharper questions
Understanding organizational dynamics
Knowing which trade-offs matter
Recognizing subtle interpersonal signals
Framing choices so leaders can act
These are the equivalent of plate discipline, defensive positioning, or pitch sequencing—skills that don’t show up easily in the box score but change the outcome of games.
Humans who can integrate AI into their process, and who excel in these upstream dimensions, will generate far more “wins” than those who compete directly with AI at the task level.
3. WAR in Knowledge Work Becomes Contextual, Not Universal
Baseball WAR is intentionally abstracted from situation-specific value. In organizations, context rules everything.
Two employees with identical AI-boosted technical outputs might produce very different “wins” depending on:
their credibility with stakeholders,
their understanding of the culture,
their ability to shape a narrative people believe,
their judgment under pressure,
their ability to align people behind a decision.
In other words, WAR in knowledge work becomes less about what you can produce and more about how your presence improves the performance of others.
Once AI handles baseline execution, value shifts to influence, coordination, decision design, and leadership.
4. AI Raises the Floor Faster Than It Raises the Ceiling
One of the overlooked implications of current models is that they compress the bottom of the performance distribution. Someone who was previously a “replacement-level” communicator can now produce work that looks like mid-level output with the right prompts and review process. Teams become more consistent, variance shrinks, and minimum standards rise.
But raising the ceiling—elite-level insight, relationship-building, strategic judgment—remains harder. These are built on lived experience, emotional intelligence, and tacit knowledge that’s not easily codified.
In this world, the WAR calculation increasingly rewards:
the ability to orchestrate teams of people and models,
the ability to discern which AI outputs are directionally useful and which are misleading,
the sensitivity to know what the client, boss, or audience actually needs (not just what they asked for).
Ceiling-raising human abilities become the real differentiators.
5. WAR as a Management Tool: Designing Roles for Maximum Value
For leaders, the WAR framing becomes a practical lens:
What tasks are we assigning people that AI can now do at replacement level?
Where do we genuinely need human insight versus competent execution?
How do we pair people and models so that the combination produces more wins than either alone?
Where are we paying for human time but getting replacement-level output?
Organizations that embrace this lens can redesign jobs to emphasize the parts of work where humans still create above-replacement value—and strip away the rote tasks that dilute it.
6. Individuals Must Manage Their Own WAR Curve
For workers, the question becomes personal:
What are the activities where I deliver value that AI cannot replace—or where my ability to use AI allows me to deliver exponentially more?
This encourages people to:
deepen uniquely human skills,
become skilled orchestrators of AI systems,
cultivate judgment in domains where pattern recognition isn’t enough,
build relationships that compound over time,
and specialize in tasks where stakes, ambiguity, or politics dominate.
Your WAR is no longer measured just by your individual contributions, but by how much better your team performs because you’re in the lineup.
7. The Ultimate Lesson: WAR Measures Contribution, Not Activity
The most important implication of the analogy is this: WAR doesn’t care how busy you are. It only cares about outcomes that move the team toward more wins.
As AI takes on more activity—the volume of work, the visible effort—the differentiators become:
insight,
clarity,
persuasion,
trust,
and the ability to make others better.
This is precisely where human leverage lives in an AI-rich environment.
In the End: WAR Reframes the Future of Knowledge Work
The rise of AI doesn’t reduce the importance of human work. It refocuses it. A rising replacement baseline forces individuals and organizations to redefine what counts as real value.
Just as in baseball, those who understand their own WAR—where they truly add wins—will thrive in a landscape where competent execution is increasingly automated. The question is no longer “Can you do the task?” but “Does your presence meaningfully improve the outcome beyond what’s generally available?”
In the AI era, that’s the new box score.
The WSJ charts on infrastructure contraints really ground the AI hype in reality. While GPT-5.1 shows meaningful improvments, the physical bottlenecks around transformers and energy production are sobering. The WAR framework for knowledge work is brilliant and captures how value is shifting upstream to judgment and context rather than execusion.