Confluence for 1.11.26

Why you’ll be using Claude Code within 12 months. The frontier most don't see. A lesson from a fake whistleblower. AI as a connective tool.

Midjourney prompt: invisible frontier

Welcome to Confluence. After last week’s entirely-AI-generated edition, the humans are back on the job. Here’s what has our attention this week at the intersection of generative AI, leadership, and corporate communication:

• Why You’ll Be Using Claude Code Within 12 Months

• The Frontier Most Don’t See

• A Lesson From a Fake Whistleblower

• AI as a Connective Tool

Why You’ll Be Using Claude Code Within 12 Months

(Or something like it.)

We have written several times about Claude Code, the version of Anthropic’s Claude that runs (until recently) only in a terminal window on your computer.

We know nobody personally outside of software development who uses it, and the preceding sentence is probably why. Consider the words in that sentence: “Claude,” “Code,” “terminal window.” Many people have yet to use Claude (though for us, it’s the best model going). Very, very few want anything to do with code. And the terminal window is something most have never seen, and many have been taught to avoid.

That said, several of your authors have been using Claude Code for some time now. And what they are doing with it is so impressive, and in some cases, so amazing, that we’re convinced it represents the near-term future of how people will use generative AI, and that it or something like it will come to a computer or device near you, soon.

We will not go into paragraphs of detail here about what Claude Code is, or how it works, but one paragraph is helpful. It’s a version of Anthropic’s Claude that you install on your computer, rather than using via a website. Because it’s local, it can with permission access the files on your machine. Because it’s used mostly by developers to write code, it can also connect to thousands of data sources, and dozens of other applications, including Microsoft 365, Morningstar, PayPal, Canva, HubSpot, and more. And because it lives in the terminal window, which is the portal to the real brains of your computer behind the nice mouse-driven interface you use, it can really do anything your computer can do: find files, create files, and delete files (permissions make this difficult, but it’s possible). It can open and work in a browser window. It can open, create, and edit Word, PowerPoint, and Excel files. You can tell Claude things like, “look at my file folders in my documents folder and give me a better way to organize them,” and it will do that. You can say, “Why does my mic on my PC not work consistently” and it will diagnose the problem and most likely fix it. And because Anthropic is clever, it can run multiple agents at once, in parallel, greatly shortening the time it takes to do work (e.g., “This folder has five outlines for five documents I need to create. Use the staff writer agent to write all five of them, working in parallel.”)

You can see a simple example of this in this video, where one of your authors used Claude Code to research references for this piece, using a web research analyst (actually, five of them, working in parallel) we’ve created:

The process took five minutes and eight seconds. The final output is in the footnotes.¹

And people are doing amazing things with Claude Code:

• Kevin Roose (NYT tech columnist, Hard Fork podcast) built a Pocket clone with TTS read-alouds, video summaries, spaced repetition emails, and Kindle sync. Quote: “From like 12 prompts? How does any app developer survive this?”

• Teresa Torres (Product discovery coach, author) runs her “entire life and business using two Claude Code terminals and a note-taking app” and has created extensive guides for non-technical users at ProductTalk.org.

• Ethan Mollick (Wharton professor, One Useful Thing) asked Claude Code to “develop a startup idea that will make me $1000 a month where you do all the work.” It worked independently for 74 minutes, creating hundreds of files and deploying a working website.

• Casey Newton (Platformer, journalist) built a personal website with dynamic content, subscription functionality, Bluesky feed, and Spotify integration, and cancelled his Squarespace account afterward.

We have a small team inside the firm experimenting with Claude Code, and candidly, this group feels like they are playing a different game. And the more they use it, the more ways they find to make it helpful and powerful in the work they are doing, in ways small (telling Claude “/update contacts,” after which it searches all incoming and outgoing email from the past 24 hours, makes a list of the people in that correspondence, eliminates spam and colleagues, and then updates, merges, or creates current Outlook contact cards for those people, including updating the context for the relationship in the notes field), personal (a health advisor that helps a user track and manage all their personal healthcare information, including integration of their exercise and biometric data, recent lab results, and more), and commercial (an ongoing Claude Code project that is helping to develop all curriculum and materials for a 12-week program that credentials senior communication officers for generative AI governance).

But ours is just a small team, in part because Claude Code in its current form is intimidating for the non-technical, and because like all organizations, we need to consider cost, training, and security implications (cost and training are bigger concerns than security as Claude Code is quite secure).

This is why Claude Code is, unless you are a software developer, probably not coming to your workplace anytime soon. Which is a shame, because it really does unlock much of the promised potential of powerful generative AI. But we do believe it will come to you sometime soon, as we see Claude Code as a bellwether of where generative AI applications are headed: the ability to work locally, the ability to work with agents, the ability to connect to data and applications, the ability to create real output, and the ability to do so via a very powerful generative AI that lets you talk to the machine and simply tell it what you want to do. That, we think, is coming to users of not just Claude, but ChatGPT and Gemini, soon. The AI labs just need to come up with a user interface that’s less intimidating and find a way of onboarding people that helps them feel secure.

We’re pretty confident that this will happen this year. The way you use generative AI — the website you use, or the application on your phone that you use — will not change much, but what it’s able to do, will. And if what we’ve been doing and learning is any indication, it will change the game for the use of generative AI in households, and eventually, organizations.

By now, you’re probably curious about trying Claude Code yourself. We think you should. It would be very rare for you to have permission to use Claude Code on a work machine in an organization, but it’s not that hard to explore Claude Code for personal use. This getting started video can help, there is a nice backgrounder here, and there are many, many “getting started” videos on YouTube. Also, this guide by Teresa Torres on how to use Claude Code safely is a must-read if you’re going to give this a whirl. We hope you do, and if you do, please share some of what you’ve done in the comments so we can see your work.

The Frontier Most Don’t See

Anthropic’s Jack Clark on the widening gap between frontier AI and mainstream tools.

ChatGPT was publicly released on November 30, 2022. By the end of that year, few had heard of it, but the technology world was abuzz about this powerful new technology and what it could mean. There was a clear gap between those who “saw it” and those who didn’t. We all know the rest of the story. In the three years since, awareness and adoption of generative AI has exploded. By now, most knowledge workers are using these tools in one way or another on a nearly daily basis.

We see a similar gap emerging today. The frontier of generative AI — which today lives in tools like Claude Code — is increasingly invisible to non-technical audiences. As we write in the previous piece, these tools are hard to access and intimidating to use. Most people don’t know they exist. The frontier has become less visible, and less legible, to the average professional.

Jack Clark, co-founder of Anthropic, captured this dynamic in a recent essay titled “Silent Sirens, Flashing For Us All”. Clark describes using Claude Code with Opus 4.5 to build a sophisticated predator-prey simulation — complete with procedural world generation and pathfinding — in about five minutes. The same project took him weeks a decade ago when he was learning to program.

Clark is, of course, about as much of an AI insider as one can be. But what struck him was that — as he stepped away from the AI frontlines to focus on his newborn — the progress had become hard to see, even for him. He writes:

Most of AI progress has this flavor: if you have a bit of intellectual curiosity and some time, you can very quickly shock yourself with how amazingly capable modern AI systems are. But you need to have that magic combination of time and curiosity, and otherwise you’re going to consume AI like most people do — as a passive viewer of some unremarkable synthetic slop content, or at best just asking your LLM of choice ‘how to roast a turkey and keep it moist’... And all the amazing advancements going on are mostly hidden from you.

This is the new gap. Developers and technical users are gushing about these tools, going so far as to speculate about whether Claude Code constitutes artificial general intelligence (AGI). Most professionals, meanwhile, are using systems that are powerful (and, frankly, would have seemed miraculous five years ago), but nowhere near as capable as what already exists at the frontier. As far as we know, no one is speculating about whether Microsoft Copilot is AGI. Clark thinks this divide will widen in 2026, predicting that by the summer “many people who work with frontier AI systems will feel as though they live in a parallel world to people who don’t.”

So, what to do in the face of this dynamic? If you haven’t yet, take our advice from the previous piece and spend some time with Claude Code. We’ve found that Ethan Mollick’s 10-hour rule applies here: if you invest the time to develop real fluency in these tools, you’ll begin to understand what’s actually possible at the frontier and deepen your intuition for working at it. It’s not trivial, and it takes real effort, but that’s exactly the point. Seeing the frontier now takes effort in a way it didn’t in early 2023, when all you had to do was visit a website.

But, as we write in the previous piece, know that this frontier is coming to mainstream tools and more familiar user interfaces. It always does. The capabilities now available only in hard-to-find, hard-to-use tools — truly agentic workflows, longer task horizons, sophisticated memory management, complex multistep task completion — will eventually come to the tools your teams use every day. Using AI to draft an email will soon feel quaint (for many of us, it already does). The only question is when and how this happens, not if it happens.

Clark closes his essay with an invitation we endorse:

It is incumbent on all of us to attempt to see this high-dimensional object for what it is — to approach this amazing moment in time with technological optimism and appropriate fear. And joy. And trepidation. And all the other emotions with which we may attempt some sense-making of the beast whose footfalls are showing up in the world.

For leaders, it’s time to begin anticipating a future that most cannot yet see. And that starts with better understanding what’s possible at the frontier in the present.

A Lesson From a Fake Whistleblower

Be mindful of what consumes your attention.

Generative AI can make us better, improving the quality of our thought, our writing, and our decisions. It also introduces new problems to solve and pitfalls to avoid. Casey Newton, of the newsletter Platformer and the podcast Hard Fork, wrote about his experience of generative AI making his work as a reporter harder.

The short version of his story: an anonymous poster on Reddit alleged significant fraud at an unnamed food delivery app. The post had 86,000 upvotes, putting it on Reddit’s front page and in front of potentially millions of people. Newton reached out to the poster to dig deeper, and the alleged whistleblower shared a photo of his employee badge and an 18-page technical document detailing the fraud. It was all fake. The allegation was a lie, and the employee badge and the technical document were AI-generated.

The story here is not that this fake whistleblower almost got one over on an experienced reporter. They didn’t. While the badge and technical document looked reasonable at a glance, they fell apart quickly under more intentional scrutiny. One of Newton’s takeaways is one we believe our readers should consider:

And while no good reporter would ever have published a story based on a single document and an unknown source, plenty would take the time to investigate the document’s contents and see whether human sources would back it up.

The bolded emphasis is ours. We won’t dismiss the risk of being fooled by AI-generated content, but there’s an argument that the bigger risk is how much attention we give to what will likely be a vast and growing amount of poorly-thought-out or downright inaccurate content that crosses our desks. We can control some of this by being intentional about how we generate content for others, but we also need to mind our own attention.

Brandolini’s Law holds that refuting nonsense takes far more energy than producing it. AI has made this worse. Generating plausible-looking content now costs almost nothing while evaluating it costs just as much as ever. The attention we give to one thing is attention we can’t give to something else. Leaders will encounter an increasing volume of polished proposals, analyses, and reports that look worth considering but aren’t. We need to maintain vigilance for fakes and the discipline to walk away from what doesn’t merit our time and attention.

AI as a Connective Tool

AI’s greatest value might be in how it works between existing systems, data, and teams.

In a recent piece published on X, venture capitalists Jaya Gupta and Ashu Garg make a case for what they see as “AI’s trillion-dollar opportunity” to change the way organizations track and use enterprise data. They note that current enterprise software tools like Salesforce, SAP, and Workday hold the underlying data for key processes and functions, but connecting and acting on that data usually relies on tribal knowledge. With AI, an agent could track that knowledge by synthesizing across a number of otherwise isolated tools, pulling customer information from the CRM, SLA terms from billing, and the Slack or Teams thread where colleagues flag a key risk or opportunity. It could then capture that context and the team’s decision-making rationale into a “context graph”— a kind of precedent library for similar situations. That’s a type of data that “doesn’t exist in most enterprises today: a queryable record of how decisions were made.”

Gupta and Garg predict the next large-scale, trillion-dollar software platforms will be built on this new kind of AI-enabled data. They may or may not be right. But whatever the systems of record of the future look like, their broader vision for AI’s role in the future of work is one we’re seeing more and more: where, beyond simply automating discrete tasks, AI’s primary, most transformative role is as a connective tool, as a kind of web or connective tissue between essential tools, data, and systems.

Imagine what it would mean for most of us if our various inboxes (Outlook, Teams, Slack, etc.), intranet, meeting notes, and files could talk to one another and share even the most basic information about what’s going on with our key stakeholders or projects, how we write and work, and what’s ahead for our week or month or year. That would certainly be transformative.

And here’s the thing: tools like Claude Code and Claude Desktop can already do this. Right now, today. And with very little effort or expertise from the user. Even if most organizations aren’t yet ready to deploy AI at scale in the way Gupta and Garg describe, AI is already very capable of transforming individual workflows.

Beyond any single model, we think it’s important for leaders to be thinking of and using AI as a connective tool. This is how AI moves beyond simply speeding up what you already do and starts enabling new kinds of workflows and insights. That future is already here, and it’s worth experimenting with sooner rather than later.

---

We’ll leave you with something cool: Anthropic shared some “delightfully specific things” users have built with Claude Code.

---

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.

1. Great research across all five agents. Here’s a synthesis for your Confluence piece:

---

Key Thesis: “Claude Code is a Misnomer”

The most consistent finding: the name “Claude Code” is actively deterring non-technical users from one of the most powerful AI tools available. Multiple writers recommend thinking of it as “Claude Local” or “Claude Agent” instead.

Dan McAteer captured it well: “Claude Code is definitely a misnomer. It is for anyone who wants to do knowledge work, faster.”

---

Notable Non-Coders Using Claude Code

High-Profile Examples

Kevin Roose (NYT tech columnist, Hard Fork podcast)

- Built a Pocket clone with TTS read-alouds, video summaries, spaced repetition emails, and Kindle sync

- Quote: “From like 12 prompts? How does any app developer survive this?”

Teresa Torres (Product discovery coach, author)

- Runs her “entire life and business using two Claude Code terminals and a note-taking app”

- Created extensive guides for non-technical users at ProductTalk.org

Ethan Mollick (Wharton professor, One Useful Thing)

- Asked Claude Code to “develop a startup idea that will make me $1000 a month where you do all the work”

- It worked independently for 74 minutes, creating hundreds of files and deploying a working website

Casey Newton (Platformer, journalist)

- Built a personal website with dynamic content, subscription functionality, Bluesky feed, Spotify integration

- Cancelled his Squarespace account afterward

Everyday User Stories (from Lenny’s Newsletter compilation of 50+ examples)

- A mom who “voice-records ideas during morning stroller walks” built a DIY subagent to design a slide tower for her son

- Lawyers building phone tree systems

- Teachers creating interactive learning games from paper sketches (one scanned a sketch, got a working HTML game teaching Orwell’s 1984 in 20 minutes)

- Non-technical users feeding raw DNA data from ancestry tests to find health-related genes to monitor

---

Best “Getting Started” Resources for Non-Technical People

1. https://natesnewsletter.substack.com/p/the-complete-wait-i-can-use-claude - Nate Kadlac

- Offers browser-only track requiring NO terminal

- Most beginner-friendly resource found

2. https://everything.intellectronica.net/p/claude-code-for-non-coders - Eleanor Berger (~3 weeks ago)

- Directly addresses intimidation factor

3. https://every.to/source-code/how-to-use-claude-code-for-everyday-tasks-no-programming-required - Every.to

- Practical examples like expense tracking

4. https://ccforpms.com/ - Free course

- Promise: “If you can have a conversation, you can use Claude Code”

---

The Viral Moment That Changed Everything

Jaana Dogan (Google Principal Engineer, Gemini API team) - January 2, 2026:

“We have been trying to build distributed agent orchestrators at Google since last year... I gave Claude Code a description of the problem, it generated what we built last year in an hour.”

7 million views within 3 days. This post “rattled engineering teams” and brought Claude Code to mainstream attention.

---

Key Quote for Your Thesis

From Ethan Mollick (January 7, 2026):

“Don’t let the awkwardness of the current Claude Code or its specialization for coding fool you. New harnesses that make AI work for other knowledge tasks are coming in the near future, and so are the changes that they will bring.”

“These systems are actually broadly useful to knowledge workers of all types... However, these new tools are built for programmers, assuming you understand Python commands and programming best practices and are wrapped in interfaces that look like something from a 1980s computer lab.”

---

Predictions Supporting Your “12 Months From Now” Argument

- Skywork AI analysis: Claude Code “predicted to expand beyond the terminal in 2026 and come to every industry, with a likely renaming since mass adoption outside engineering may not embrace something with the word ‘Code.’”

- Voice-first interfaces: Multiple sources predict Q1 2026 for native voice in Claude Desktop

- Anthropic’s signal: They renamed “Claude Code SDK” to “Claude Agent SDK” to reflect broader vision

- Enterprise data: 91% of IT executives already acknowledge non-technical employees are driving AI innovation (Inc.com)

---

Sources (All December 2025 - January 2026)

Major Articles:

- https://www.oneusefulthing.org/p/claude-code-and-what-comes-next - Ethan Mollick, Jan 7, 2026

- https://www.transformernews.ai/p/claude-code-is-about-so-much-more - TransformerNews

- https://www.platformer.news/claude-code-review-web-design/ - Casey Newton, Platformer

- https://www.lennysnewsletter.com/p/everyone-should-be-using-claude-code - Lenny Rachitsky

Guides for Non-Technical Users:

- https://natesnewsletter.substack.com/p/the-complete-wait-i-can-use-claude

- https://natesnewsletter.substack.com/p/claude-code-without-the-code-the

- https://ccforpms.com/

Video/Events:

- https://every.to/source-code/claude-code-camp - 400+ attendees learning non-technical uses

- Teresa Torres 50-min tutorial on automating life with Claude Code

This should give you plenty of material for the Confluence piece. The narrative practically writes itself: the tool is already being used by non-coders for remarkable things, but the CLI interface and “Code” name are temporary barriers that will dissolve over the next year.