I’m increasingly finding Claude useful to tell me how to best use Claude.

I’m working on a project right now with a bunch of colleagues, and one of the limitations of Claude, even the Pro version, is that there is no option to invite other Claude users to collaborate on a project. In order to do that, you have to upgrade everyone to a Claude Team or Enterprise license — which may not be feasible for cost or logistical reasons.

Our specific use case: a few of my Markup AI colleagues and I are vibe coding an app together in Replit (I can’t wait to tell you about it when it’s further along), and we are doing a bunch of the design work in Claude. While Replit is collaborative, Claude is not. All of the initial work on the project was done inside one person’s Claude instance and wasn’t shared.

So after a lot of back and forth with Claude, here’s the workaround we landed on.

---

The Core Idea

Use Google Drive as your shared brain. Use Claude’s Custom Instructions as the mechanism that loads it at the start of every conversation.

Each person keeps their own Claude Project. But instead of context living in isolation inside each person’s account, it lives in three shared Google Docs that every team member’s Claude reads from automatically.

---

The Three-File Architecture

Create a folder in Google Drive — something like [Project Name] / Claude Context — and put three Google Docs in it:

ProjectContext — everything Claude needs to know about the project. What it is, key decisions, the framework you’re working within, roles, terminology, things that have been explicitly decided and shouldn’t be revisited. If you hired a really smart person tomorrow and handed them one document, this is that document.

WorkingConventions — how you want Claude to behave. Communication style, output format, tools and tech stack, things Claude should never assume, any role-specific preferences individual team members have.

ChangeLog — a running dated log of decisions, newest at the top. When the full ProjectContext hasn’t been updated yet, this captures what’s new. It’s the lightweight sync mechanism that keeps things from getting stale between major updates.

Use consistent, searchable names — we use [ProjectName]_ProjectContext, [ProjectName]_WorkingConventions, and [ProjectName]_ChangeLog. Exact naming matters because Claude searches for them by name.

---

The Custom Instructions

Every team member pastes this into their Claude Project settings (Project → Settings → Custom Instructions):

You are working on [PROJECT NAME].

At the start of every conversation, before responding to anything
else, search Google Drive for these three files and read all of
them in this order:

1. [ProjectName]_ProjectContext — the full framework, key
   decisions, architecture, and all settled decisions.
   This is the source of truth.
2. [ProjectName]_WorkingConventions — how to work with this
   team: communication style, output preferences, and
   what not to do.
3. [ProjectName]_ChangeLog — the running log of recent
   decisions. Read this to catch anything decided since
   the context doc was last updated.

These files are in Google Drive under [Folder Name]. Do not
rely on training data or assumptions about this project. Always
ground responses in the current versions of these documents.
After reading them, proceed without narrating that you have
done so — just apply the context.

The “proceed without narrating” line matters. Without it, Claude opens every response with a paragraph about how it just searched Drive and found the files and read them all carefully. Gets old fast.

You can layer personal preferences on top of the shared core. The shared instructions load the shared context; any personal additions shape how Claude works with you specifically.

---

The Google Drive Connector

Each team member needs the Google Drive connector active in their Claude instance, authenticated to the Google account where the files live. In claude.ai, this is under Settings → Integrations.

Here’s the thing that tripped us up: the connector authenticates to a single Google account per user. If your project files live in a Google Workspace account and someone’s connector is pointed at their personal Gmail, they won’t find anything. Make sure everyone connects to the right account before you waste time debugging. The workaround here (yes, the workaround nested inside the workaround) is to use Google Drive to sync a version of your files locally one of your two Drives, then use the FileSystem connector instead of the Google Drive connector, and make sure you note that clearly in your version of the Instructions.

---

Verify It’s Working

Before doing any real work, each team member should run this as their first message in the new Project:

Context load test. Search Google Drive for
[ProjectName]_ProjectContext, [ProjectName]_WorkingConventions,
and [ProjectName]_ChangeLog and read all three. Then answer
these questions from what you read — no assumptions, only
what's in the documents:

[Insert 3-5 questions whose answers are clearly in your
ProjectContext]

After answering, report: which files did you find, and were
there any you couldn't locate?

Design the questions so a correct answer requires actually reading the content — not something Claude could guess from the project name alone. If it gets the answers right and reports all three files found, you’re good.

---

Keeping It Current

This only works if the documents stay current. Two habits make that sustainable without it becoming a chore:

The ChangeLog is the low-friction entry point. Decision gets made? Throw a dated one-liner into the ChangeLog. Don’t worry about updating the full ProjectContext every time — that can wait for the weekly pass.

The weekly harvest prompt. Once a week, run this in your Project:

Weekly context harvest. Search my recent conversations in
this project from the past 7 days. Identify any decisions,
framework changes, or structural updates that were reached
in those conversations but are not yet reflected in the
three context files in Google Drive.

For each item you find, give me:
1. What was decided or clarified
2. Which file it belongs in
3. The exact text to paste in, formatted and ready to copy

Group your output by file so I can do three copy-paste
operations total. If nothing new was decided this week,
tell me that explicitly.

Post the output to a shared Slack channel. One person — we call them the context owner — consolidates and pastes into the Drive files. Everyone’s Claude picks up the updates on the next conversation.

---

What This Does and Doesn’t Do

It does:

• Give every team member’s Claude the same foundational context, automatically
• Keep that context current through a lightweight weekly process
• Require zero engineering, no special plan, no new tools
• Scale easily — adding a new person is just “set up a Project and connect Drive”

It doesn’t:

• Enable real-time collaboration (Claude still can’t see what’s happening in someone else’s conversation)
• Write decisions back to Drive automatically (the connector is read-only; you still copy-paste updates by hand)
• Share personal memory across team members

---

Quick-Start Checklist

1. Create a Google Drive folder: [Project] / Claude Context
2. Create three Google Docs: [Project]_ProjectContext, [Project]_WorkingConventions, [Project]_ChangeLog
3. Write the content for each — ProjectContext first, that’s where the real work is
4. Each team member: create a Claude Project, paste the Custom Instructions, connect Google Drive to the right account
5. Each team member: run the test prompt to verify all three files load
6. Designate a context owner and pick a weekly harvest day
7. Set a recurring Friday reminder

It’s not a perfect substitute for native collaboration — Anthropic, if you’re reading this, please just build multi-user Projects — but it’s a surprisingly effective workaround that costs nothing and takes about an hour to set up.

---

Have you found other ways to get Claude to collaborate across multiple accounts? I’d love to hear what’s working.

One of the hardest things about being a CEO in the AI era isn’t the technology itself — it’s the firehose of information about the technology. There’s so much being written about AI right now that it’s almost impossible to separate the signal from the noise. Hot takes, doomsday predictions, breathless hype, vendor pitches dressed up as thought leadership — it’s exhausting.

So I thought I’d do something useful and share periodically a curated basket of the most interesting reading I’ve done. Think of it as the reading list I’d hand to a fellow CEO who said, “I know I need to get smarter about AI — where do I start?”

This first batch is a bit of a catch-up, but it’s a strong starting point. Here goes.

The Big Picture: Hope and Fear from the Same Person

Start with these two essays from Dario Amodei, co-founder and CEO of Anthropic (the company behind Claude). Read them as a matched pair — they’re essentially the optimist and pessimist cases from someone who understands AI as deeply as anyone on the planet.

• Machines of Loving Grace (October 2024) — Amodei’s case for AI’s transformative upside. How AI could compress a century of medical progress into a decade, address poverty, strengthen democracy. Long but worth every minute. This is the essay that made me think, okay, this really is different from every other tech wave.
• The Adolescence of Technology (January 2026) — The companion piece, and it’s sobering. Amodei confronts the real risks: national security, economic disruption, democratic erosion. The title comes from the movie Contact — how does a civilization survive its own technological adolescence without destroying itself? A question worth sitting with.

The Software Factory: What AI Is Doing to Developers

This cluster of posts is the best thing I’ve read on how AI is actually changing the work of building software — and by extension, all knowledge work.

• The Five Levels: From Spicy Autocomplete to the Dark Factory — My friend Dan Shapiro (CEO of Glowforge, who also inspired our Chatgipity platform) created a memorable analogy framework modeled on the five levels of autonomous driving. Level 0 is “spicy autocomplete.” Level 5 is the “Dark Software Factory” where AI builds software autonomously. It’s the clearest mental model I’ve seen for understanding where we are and where this is headed.
• Michael Bernstein’s counterpoints — My colleague Michael offered some thoughtful shaping of Dan’s framework that’s worth reading alongside it. Not disagreement so much as nuance around the edges.
• Simon Willison’s amplification — Willison, one of the most respected voices in the developer community, picked up Dan’s post and extended the thinking around automation. If you follow only one technical blogger on AI, it should probably be Simon.
• Something Big Is Happening by Matt Shumer — Related to the above but pushes further into societal territory. Shumer starts getting into the employment implications of what happens when AI can do more and more of what knowledge workers do today. Provocative and worth reading even if (especially if) you’re skeptical.

The Longer View: Employment, Society, and the Pace of Change

• This thread is a more rational, long-term view of the employment question. Less alarmist than most of what you’ll find, and grounded in historical precedent. A useful counterweight if the Shumer piece made you nervous.
• 2028 GIC — This one is fiction, and it’s poignant. It imagines the societal impacts we could see from AI in the near term. I’ll be honest — as a piece of speculative writing, it hit me harder than most of the non-fiction. If nothing else, it reinforces my view that things are happening really quickly, probably more quickly than both individual humans and human institutions like governments will be able to keep pace and react to. That’s not a reason to panic. But it is a reason to pay attention.

More to Come

I’ll do these roundups periodically. If you’re reading something great about AI that you think I should see, send it my way.

AI is going to destroy a lot of jobs. Let’s just start there. White collar jobs. Desk jobs. The kind of jobs where your primary output is information, analysis, or words on a screen. This isn’t speculation — it’s already happening.

But here’s the thing: the world has survived every major technological disruption in history. When the power loom arrived in the early 1800s, hand weavers rioted — literally smashed the machines — because they were certain it was the end of work. It wasn’t the end of work. It was the end of that work. New work emerged that no one could have predicted, like, oh say, the commercial mass-produced clothing industry, which had even more jobs on the other side of its creation, just different jobs.

The problem is that when you’re standing in the middle of a disruption, all you can see is what it’s going to destroy. You can’t yet see what it’s going to create. That’s the cognitive trap, and we’re deep in it right now.

History Lessons: The Good, the Bad, and the Fast

Think about how poorly globalization landed in terms of job displacement. It played out over 30 years, and we still didn’t have adequate retraining programs, social safety nets, or political will to manage the transition well. Entire communities were gutted while politicians debated.

We did reasonably well with the computing revolution — probably because it created visible new industries and job categories in parallel with the ones it displaced, and because the transition was gradual enough for institutions to adapt.

But this one? This one is going to be faster. Dramatically faster. And it’s arriving in a political universe that is less equipped to handle it. Governments can barely agree on basic tech regulation (or let’s be real, much of anything), let alone coordinate workforce transitions. The policy infrastructure to manage AI-driven displacement doesn’t exist yet, and the pace of AI advancement is not waiting for Congress to catch up. That gap — between the speed of the technology and the speed of institutional response — is the real danger zone.

The Window Is Open, But It’s Closing

Elena Verna nailed it in a recent post: “AI won’t take your job. Being complacent about what’s happening around you will.” She argues there’s a short window to get radically ahead by going AI-native, and she’s right. That window is open now. It won’t be open forever.

The Wall Street Journal ran a piece recently on what young workers are doing to “AI-proof” themselves. Some are pivoting to sectors less exposed to automation. Others are doubling down on AI skills within desk-job sectors. Both are rational strategies. But here’s the main point I want to make:

AI won’t take all jobs. We will still need software developers and paralegals. But we will need far fewer of them. And if you’re worried about not getting one of those remaining jobs, not keeping the one you have, or not getting promoted as fast — your best recourse is to get really, really good at using AI.

I mean really good. Not “I ask ChatGPT questions sometimes” good.

What “Really Good” Actually Looks Like

You MUST go beyond using AI as a fancy search engine or writing assistant. Here’s what I think the bar is:

• You should be building Custom GPTs or Gems for your specific workflows
• You should use multiple LLMs and understand the practical differences between Claude, GPT, and Gemini
• You should be using Claude Cowork, the Claude Chrome Extension, and Perplexity Computer to understand how they boost application productivity
• You should experiment with the AI browsers — Atlas and Comet — and see how they change your relationship with the web
• You should vibe code a couple of applications using Lovable or Replit
• You should figure out what OpenClaw is and why it’s the hottest AI agent tool of 2026

You should be talking about all of this on your résumé, on your LinkedIn profile, and in your interviews — with concrete examples, not buzzwords. Think about it this way. If you were the hiring manager for whatever job you’re interviewing for, and you have two finalist candidates, all else equal, which one will you hire – the one who makes a joke or snarky comment about AI, or the one who very succinctly explains how she used it to save her prior employer money or created an application to help her elderly parents manage their medication more effectively?

And if all else fails and you don’t get the job you wanted as a management consultant or paralegal, well, at least you’ll have learned a ton about how to use AI. That skill set will come in handy finding something else. No matter what the purpose and the use case, this is a good investment of your time.

A Note to Employers

I want to end with something I feel strongly about: please don’t fire all your junior people just because Claude can do their job today.

Someday, you’re going to need a batch of new not-junior people. And the way you get those people is by training junior people. Junior people have to learn by doing — not only by being clever at writing prompts. You can’t skip the apprenticeship phase of a career and expect to have seasoned leaders in five years.

Yes, white collar organizations are going to look different. More like stovepipes and less like pyramids. Fewer layers, fewer people per layer, with AI handling much of the work that used to require entry-level headcount. The organizational chart of 2030 won’t look like the one from 2020. That’s okay. But we have to be intentional about not hollowing out the pipeline of human talent in the process.

In May 2023, Fred Wilson wrote a post about being approached by a company that had trained a large language model on all 9,059 of his AVC blog posts. They wanted to offer a chatbot called “Ask Fred.” He said no thanks.

His reasoning was sharp. He’s fine with anyone using his content to train AI. He put a Creative Commons license on his blog from the start. But he didn’t want a bot pretending to be him. The whole point of his blog is the humanity — the daily conversation, the thinking out loud, the relationship with readers. A chatbot that mimics Fred Wilson isn’t Fred Wilson. It’s a parlor trick.

I read that post and agreed with Fred completely. An AI version of me that replaces me? No interest.

But an AI version of me that helps me be more productive while keeping my voice and point of view? That’s a different story entirely.

Building MattBot

MattBot is a custom agent I built inside Chatgipity, our company’s internal AI platform. It’s not a gimmick. It’s one of the most useful tools I’ve ever created.

Here’s what’s under the hood:

• My complete body of writing. Every blog post from 20+ years of StartupCEO.com. My three books — Startup CEO, Startup Boards, and Startup CXO. My eBooks. Podcast transcripts from over 200 episodes. Conference talks.
• My email and Google Drive. So it has context on what I’m working on, who I’m talking to, and what’s current.
• A detailed map of what to include. I didn’t just dump content in. I created a structured guide for the agent — what topics I cover, what my positions are, how I frame arguments, what language I use and don’t use.

The result is an agent that doesn’t replace me. It accelerates me. It knows how I think, how I write, and what I’ve already said about many topics in entrepreneurship and leadership.

How It Saves Me Time

Let me give you some real examples.

A college kid I used to coach in baseball reached out recently and asked to interview me for his freshman entrepreneurship class. I always say yes to these — it’s important to me to help young people who are curious about building things. But historically, that’s a 90-to-120-minute commitment: time to think about the questions, time to write thoughtful responses, time for follow-ups.

This time, I asked him to send me the questions ahead of time. I fed them to MattBot. In 30 seconds, I had draft responses that were substantively right and sounded like me. I spent five minutes editing them, sent them back, and then spent 15 minutes on follow-up questions. Total time: 20 minutes instead of two hours. Same quality. Same voice. Same care.

Preparing for podcast appearances used to take me 30 minutes of reviewing the host’s questions, thinking about framing, jotting notes. Now it takes 30 seconds — MattBot drafts my talking points based on what I’ve already written and said about those topics, and I review and adjust on the fly.

Taking a generic output from any LLM and “Matt-izing” it takes 30 seconds. The upstream work on the LLM might save hours, but the only reason the final product works is that MattBot refines it with my language, my voice, and a high percentage of my long-term body of knowledge.

How the Company Uses It

This is where it gets even more interesting. MattBot isn’t just for me.

Marketing or our PR firm can hand me bullet points for an article or a LinkedIn post, and I can have MattBot do a first draft in 30 seconds. I review it, tweak it, and send it back. What used to be a multi-day back-and-forth is now an instantaneous turnaround.

Better yet, I published the agent to two colleagues in Marketing so they can use it directly. They can draft thought leadership content in my voice without waiting for me to be available. I still review everything — that’s non-negotiable — but the bottleneck is gone.

The Flywheel

Here’s what makes MattBot different from a one-time prompt or a static set of instructions: I treat it like a living system.

Every time I review a response and make edits, I feed those edits back in and tell the agent to learn from them. Any time I write something new — a blog post, an article, an email that captures how I think about a topic — I feed it in. Every new podcast transcript goes in.

So MattBot keeps getting better over time. It’s not a snapshot of who I was when I built it. It’s an evolving representation of how I think and communicate right now.

Fred was right to say no to “Ask Fred.” But I think he’d appreciate what MattBot actually is: not a replacement for humanity, but a tool that gives me more time to be human.

A little over a year ago, my friend Dan Shapiro, CEO of Glowforge, showed me what his team had built internally with AI tools. Not one-off ChatGPT experiments. Not an AI pilot program. A real, unified infrastructure where the whole company was using AI through a common platform with shared agents, shared memory, and shared integrations. I walked away from that conversation knowing we needed to do the same thing.

A few weeks later, our VP of Technical Operations, Martin Hempstock, stood up an instance of LibreChat — an open-source AI platform that brings together all the major LLMs into a single, customizable interface. Martin had it provisioned under a new subdomain within days.  It’s changed the way we work.

The most entertaining part of this story? The name.

How a Five-Year-Old Named Our AI Platform

Our Board member Jake Heller’s five-year-old son had been hearing his dad talk about ChatGPT. Except he couldn’t quite pronounce it. What came out was “Chatgipity.” When Martin was setting up the platform and needed a subdomain for the URL, he just ran with that. And so Chatgipity was born — a name so perfectly imperfect that no one ever wanted to change it.

What LibreChat Actually Does

For those unfamiliar, LibreChat is an open-source platform (MIT licensed, 34k+ GitHub stars) that gives your organization a unified AI interface with some critical capabilities:

• Model-agnostic. You’re not locked into one LLM. We use Claude, Gemini, GPT — whatever model is best for a given task. When a new model drops, we plug it in. When one gets discontinued, we swap it out. No vendor lock-in.
• Agent creation. Anyone in the company can build, share, and publish AI agents internally. This is key. You’re not waiting for IT to build things for you — marketing builds marketing agents, product builds product agents, the CEO builds his own agents (more on that in future posts).
• App connections. Through MCP (Model Context Protocol) and integrations, our agents can talk to Google Workspace, Slack, Jira, Grafana, Confluence, Salesforce — basically the full stack of tools our team uses every day.
• Security. SSO through Google, enterprise authentication, rate limiting, moderation tools. Your data stays in your environment. For a European company like ours that takes GDPR seriously, this matters.
• Shared memory and context. Agents can share knowledge and reference the same underlying data, so you’re building institutional AI intelligence, not individual silos.

Why This Matters Waaaaay More Than Individual AI Subscriptions

Here’s the thing most companies get wrong: they buy a bunch of ChatGPT Enterprise licenses and tell people to go use AI. That’s like handing everyone a word processor and calling it a content strategy.

What you actually want is a unified framework — a common infrastructure where agents share context, where integrations are built once and used by everyone, and where the organization’s collective intelligence compounds over time.

One of the best examples at our company is the Customer Migration Tool. We needed a single source of truth for customer data across multiple systems — CRM, support, billing, product usage. Instead of building separate integrations for each system, one member of our team built one agent that pulls it all together. Built once, used by everyone.

We’ve had people build a Grafana Reporter that queries dashboards in natural language. A Calendar + Email Scanner that preps you for every meeting. A Confluence Pal that makes our documentation actually searchable. A Product Feedback agent. Even a Strength & Conditioning agent and a Berlin Public Transport agent (because why not).

Within weeks of launch, we had 48 active users sending nearly 2,000 messages. Today, it’s woven into how we work — from drafting customer emails to generating ERP RFPs to running competitive analysis.

My Take

I’ll write more about my personal agents in a future post, but I want to be clear about something: I use this every day. I’m not the CEO who mandated an AI platform and then never logs in. I build agents. I break agents. I file bugs with Martin. I write prompts. This is how I think CEOs need to operate in this era — not delegating AI to the IT team, but being a hands-on user and evangelist.

If you’re running a company and you haven’t stood up a unified AI infrastructure yet, you’re leaving an enormous amount of productivity and institutional knowledge on the table. The tools exist. They’re open source. They’re ready.

And if you need a name for yours, I know a five-year-old who might have some ideas.

My marketing team has been asking me to write more about Markup AI — what we do, why it matters, where we’re headed. And I will. I’m the CEO of a company that builds Content Guardian Agents to help enterprises scale AI-generated content smartly and safely. I love what we’re building. I believe in it deeply. And I’ll reference it when it’s relevant. I also love how we’re building things, since we are hardcore practitioners using AI to build our business at decent scale.

But if you’ve read this blog over the years — through my two decades at Return Path, my time building Bolster, and three books on startup leadership — you know that’s never been what this space is about. This blog has always been about the craft of being a CEO. The stuff that should be in a handbook somewhere but isn’t.

So no, I’m not going to turn StartupCEO.com into a company blog. What I am going to do is write a series of posts about how CEOs should be thinking about AI — because I don’t think enough of us are, and I don’t think anyone else is writing about it from the operator’s chair.

What I’ll Cover

I plan to write about the practical side of AI for leaders: how I’m using AI tools to run my company and my executive team. How AI changes board dynamics and governance. What I’m reading and what’s worth your time (consider me your AI curator-in-chief). How AI is reshaping go-to-market, product development, hiring, and decision-making. And yes, where Markup AI fits into the picture — because our work on content governance is one of the most interesting AI problems out there, and I’d be leaving out a big part of the story if I didn’t talk about it.

But this isn’t a product blog. It’s a leadership blog. That’s on brand for me.

Why This Matters: Three Kinds of CEOs

Here’s what I’ve observed over the past year. There are three kinds of CEOs right now:

1. CEOs who are actively using AI — experimenting, building workflows, changing how their teams operate, pushing the boundaries.
2. CEOs who are telling their people to use AI — they get it conceptually, they’ve mandated adoption, but they’re not in the weeds themselves.
3. CEOs who don’t get it — they think this is someone else’s problem, or that it will sort itself out, or that it’s overhyped.

If you’re in category one, I want to compare notes with you. If you’re in category two, I want to pull you forward. If you’re in category three…I probably can’t help you.

Why Now

I was at a Gartner CEO conference on AI not long ago, and one line keeps rattling around in my brain: “There is such a rapid pace of development that we are regularly seeing ‘obsolescence before maturity.'” In other words, by the time AI products reach early adoption, they’re already being replaced by something better.

That’s the pace we’re dealing with. And it feels familiar. I started my first company in 1999, right at the dawn of the commercial internet. This moment feels like 1995 — except everything is moving faster, the stakes are higher, and the potential impact is broader. We are living through a golden age of technology transformation, and the CEOs who lean into it are going to build the defining companies of the next decade.

I don’t say that lightly. I’ve seen hype cycles come and go. This isn’t one of them.

More to Come

So that’s what’s ahead: a series of posts, coming regularly, about AI through the lens of a CEO who’s building an AI company, running a team with AI tools, sitting on boards that are grappling with AI governance, and trying to make sense of it all in real time.

I hope you’ll come along for the ride.

We’ve been sounding the alarm on this at Markup AI for months, and now the data is catching up.

A recent Harvard Business Review audit of major LLMs just validated a critical shift: Machines are distorting brand reality. When Pernod Ricard ran the numbers, they found models miscategorizing Ballantine’s – very much a mass-market staple (and a decent scotch whiskey, though not my personal favorite, and not a high end one like Talisker or Oban) – as a prestige product.

Why is this an issue CEOs or even board members should care about?

Because the “customer journey” has fundamentally changed. With two-thirds of Gen Z and half of Millennials now using LLMs for product research, we’re witnessing the death of keyword stuffing and the birth of Share of Model.

For 20 years, we fought for the top link. But in the era of Agentic AI—where software doesn’t just search, it negotiates and buys—the rules of engagement are different. Agents don’t care about your SEO tricks. They care about your clarity.

This is exactly why we built Markup AI.

You can’t manually iterate your way to relevance in an LLM world. You need infrastructure. Optimizing for AI requires structural rigor, semantic precision, and absolute content governance. Without it, models won’t just ignore you—they’ll hallucinate your value proposition.

The future isn’t about being found. It’s about being understood.

Is your content ready to be read by a machine?

Read the full article here.

(I realize I’ve been relatively quiet since I started my new job in January…this is what I’ve been up to. It may be the most interesting job I’ve ever had…but more to come on that over time.)

I was at a recent Gartner CEO conference on AI in New York City, and one phrase keeps rattling around in my brain: “There is such a rapid pace of development that we are regularly seeing ‘obsolescence before maturity.'” In other words, by the time AI products reach early adoption in the market, they’re already obsolete.

But here’s the thing – while everyone’s racing to build the next AI breakthrough, they’re missing a massive problem hiding in plain sight. AI has fundamentally changed the content game. You can generate content faster than ever, but the risk has grown exponentially. Human review doesn’t scale at AI speeds. Companies need intelligent guardrails.

That’s why I’m excited to announce what we’ve been building at Markup AI.

The Problem No One’s Talking About

Let’s be honest about what AI does well and what it doesn’t. AI models are fantastic at content generation by the pound, but they’re not great at precision and quality. More importantly, AI can’t possibly check its own work for quality. As you know from using ChatGPT or Claude, when you push back on AI output, it responds “you’re absolutely right,” no matter what you say. That’s not exactly the kind of rigorous quality control you want for your enterprise content.

The writing assistant industry has been playing “small ball” for years – spell check, grammar, basic tone suggestions, looking at one sentence at a time as you’re writing content. They’ve ignored several elephants in the room: brand compliance, terminology management, risk management, and batch checking already-published content.

Enterprise content production and compliance is a massively fragmented problem. Companies have thousands of people who write publicly facing content and dozens of systems for authoring, storing, and publishing content on one hand; and multiple brand and policy rule books and thousands to hundreds of thousands of pieces of already-published content on the other.

We work with the world’s largest enterprises across technology, healthcare, financial services, and manufacturing – organizations that can’t afford content mistakes because the stakes are too high. One wrong sentence could mean litigation, regulatory issues, or a reputation crisis.

Markup AI: Your AI Watching Your AI

This is why we’re launching Markup AI – the first content guardian agent. Think of it as your AI watching your AI. We scan, score, and rewrite content to ensure it’s on-brand, compliant, and won’t create problems.

Fixing typos is great, but that won’t ensure your content doesn’t get you sued, fired, or create a reputation crisis. We’re making sure your content actually serves its purpose.

The name says it all: Editors markup documents. Lawyers markup documents. Writers markup their own documents. And developers regularly use markup languages and annotations. Markup AI will quickly become the standard for Content Guardian Agents in the AI space – helping companies to scan, score, and rewrite content that is consistent, authentic, and compliant, rapidly and at scale.

The Press Release announcing our launch and $27mm financing is here.

What Makes Us Different

Here’s what makes us different: we meet you where you already work. Through our API-first approach, Markup AI integrates into whatever content systems and workflows you’re using today. No rip-and-replace required.

We’re building the future of content creation – where speed doesn’t compromise safety. True AI-native solutions that tackle trust and risk head-on in an AI-first world with triggers for human-in-the-loop. Because at the end of the day, AI can write, but someone needs to ensure what it writes is actually right.

This is a Restart

This isn’t just a product launch – it’s a restart. A new AI-native platform, a new team, a new location, and a new brand. With years of linguistic engineering experience behind us in our prior business Acrolinx, the gold standard for enterprise content compliance, we’re excited to unveil our new native-AI brand and platform, Markup AI.

We’ve been at this for a while now, and we’re proud to announce that leading tech companies such as Amazon, Adobe, ServiceNow, Gainsight, Instacart, Coinbase, plus over 75 more AI-first companies have all asked for early access to our Content Guardian Agent and a number are already up and running.

The future of content is here – and it needs intelligent guardrails. We’re excited to show you what we’ve built and how Markup AI can help your organization navigate the AI content revolution safely and successfully.

Ready to see your AI watching your AI? Visit us at www.markup.ai to learn more, get an API key for free, and try out our Writer’s Playground or the API itself.

More to come on the restart journey. It’s an interesting one, full of lessons for Startup CEOs everywhere.

I’m thrilled to announce that we created mini-books (about 80 pages long and only $9-10 on Amazon) out of five of the major functional areas covered in Startup CXO: A Field Guide to Scaling Up Your Company’s Critical Functions and Teams, part of our series along with Startup CEO: A Field Guide to Scaling Up Your Business and Startup Boards: A Field Guide to Building and Leading an Effective Board of Directors.

I’ve always said that while I love all three books, in some ways Startup CXO is the best because it’s a “book of books.” While I’d still encourage all CEOs and senior executives (CXOs) to read the full manuscript, my friends and co-authors and I are happy to present these five books, now available on Amazon, for functional specialists:

• Startup CFO: A Field Guide to Scaling Up Your Company’s Finance Function
• Startup CRO: A Field Guide to Scaling Up Your Company’s Sales Function
• Startup CMO: A Field Guide to Scaling up Your Company’s Marketing Function
• Startup CPO: A Field Guide to Scaling Up Your Company’s HR/People Function
• Startup CTO: A Field Guide to Scaling Up Your Company’s Technology/Product Function

Each book has several topics in common – chapters on the nature of an executive’s role, how a fractional person works in that role, how the role works with the leadership team, how to hire that role, how the role works in the beginning of a startup’s life, how the role scales over time, and CEO:CEO advice about managing the role.

In Startup CFO, the role-specific topics Jack Sinclair talks about are Laying the CFO Foundation, Fundraising, Size of Opportunity, Financial Plan, Unit Economics and KPIs, Investor Ecosystem Research, Pricing and Valuation, Due Diligence and Corporate Documentation, Using External Counsel, Operational Accounting, Treasury and Cash Management, Building an In-House Accounting Team, International Operations, Strategic Finance, High Impact Areas for the Startup CFO as Partner, Board and Shareholder Management, Equity, and M&A.

In Startup CRO, the role-specific topics Anita Absey talks about are Hiring the Right People, Profile of Successful Sales People, Compensation, Pipeline, Scaling the Sales Organization, Sales Culture, Sales Process and Methodology, Sales Operating System, Marketing Alignment, Market Assessment & Alignment, Channels, Geographic Expansion, and Packaging & Pricing.

In Startup CMO, the role-specific topics Nick Badgett and Holly Enneking talk about are Generating Demand for Sales, Supporting the Company’s Culture, Breaking Down Marketing’s Functions, Events, Content & Communication, Product Marketing, Marketing Operations, Sales Development, and Building a Marketing Machine.

In Startup CPO (HR/People), the role-specific topics Cathy Hawley talks about are Values and Culture, Diversity Equity and Inclusion, Building Your Team, Organizational Design and Operating Systems, Team Development, Leadership Development, Talent and Performance Management, Career Pathing, Role Specific Learning and Development, Employee Engagement, Rewards and Recognition, Reductions in Force, Recruiting, Onboarding, Compensation, People Operations, and Systems.

In Startup CTO (Technology and Product), the role-specific topics Shawn Nussbaum talks about are The Product Development Leaders, Product Development Culture, Technical Strategy, Proportional Engineering Investment and Managing Technical Debt, Shifting to a New Development Culture, Starting Things, Hiring Product Development Team Members, Increasing the Funnel and Building Diverse Teams, Retaining and Career Pathing People, Hiring and Growing Leaders, Organizing Collaborating with and Motivating Effective Teams, Due Diligence and Lessons Learned from a Sale Process, Selling Your Company, Preparation, and Selling Your Company/Telling the Story.

Each of these executives is a true subject matter expert, not to mention a great friend and someone who is a lot of fun to hang out with on an executive team. I’m proud of these books and hope they’re a useful addition to the startup canon.

It’s been a big break between posts – as many of you probably know, I moved to Board Chair and left the CEO role at Bolster last summer (it’s now in the very capable hands of my friend and co-founder Cathy Hawley), and I’m now CEO of a super cool AI company called Acrolinx.  So yes, that means I went through a job search – and I found my ultimate job as a result of an inbound cold email from a headhunter!  The rich irony in that as someone who founded an executive search platform is not lost on me.

So when a good friend of mine who is also between CEO gigs and looking at several opportunities asked me the other day “why is this process so slow, and what can I do about it?” I riffed with him on the theme for a bit and thought I’d share my thinking here. 

Why Executive Searches are Slow

My top three reasons on this are pretty varied – there’s no specific theme.

1. Boards aren’t efficient hiring managers. When hiring a CEO, even the best intentioned boards can be slow to move. Frequently they operate with a search committee, and even if there’s a lead director on the search committee or even no actual search committee, by design they need to operate with a high degree of consensus. Organizing five calendars to meet with or debrief on a candidate can take weeks. And a single loud voice saying “no” or “not sure” can paralyze a board. All this is true for a CEO search but can also be true when a less experienced CEO is trying to hire a CXO and needs a lot of Board involvement in the process. At Bolster, we’ve worked on mitigating this by getting the key decision-makers aligned on search criteria at the beginning of the search, prepping interviewers, and creating a scorecard for each candidate that is visible to all decision-makers, but sometimes that doesn’t matter.
   
2. Boards and CEOs often don’t know what they want.  Whether a company is hiring a role for the first time or replacing an executive, they often get to a generic job spec but don’t actually know what they’re looking for. Not all CEOs are created equal. Not all CROs have the same core competencies. At Bolster, we developed a description of role archetypes for each C-suite or Head-or role that helps with this process (eg for a CFO, do you want an Accounting type, a Finance/Ops type, or a Deal type?). But even if a Board or hiring CEO has this level of detail down, it can still be a murky picture, trapped between the company’s past successes and failures on one side and its future needs on the other. Processes move slowly because it take a while for the picture to become less murky – circumstances around the company evolve, or people see how the company operates without this role as others pick up the slack, and therefore the needs of the role shift or come into focus. Sometimes meeting a series of candidates is the only thing that can help drive this focus, and per the first bullet above, this just takes time. If a company has a strong search partner, that may speed things up via quick presentation of calibration candidates.
   
3. There’s no precipitating crisis.  Most companies and departments, most of the time, are not in crisis. A lot of companies can operate without a given executive, even a CEO, for quite some time. Some things done (don’t) get done. Other people rise to the occasion and pick up the most important items. Or the company has hired an interim or fractional executive as a stop gap measure. Without a specific and clear sense of urgency, searches often don’t have a driving force. Sometimes there’s a precipitating crisis like a system outage or massive customer churn or the company running out of cash that can provide that driving force, but that is not the norm.

What Can You as a Candidate Do About It?

The answer is probably “not much.”  But if my own search was any indicator, I’d give you the same advice I give people internally at my company when they ask me how to get a promotion.  My answer is “start doing the job today, don’t wait to actually get the job.”  Obviously a candidate for a CEO role or any other executive role can’t actually start doing the job as an existing employee could start taking on additional pieces of work.  But there are a lot of things you can do to “act as it” and get the hiring Board or hiring CEO’s attention.  For example:

• As a CEO candidate, be a management consultant.  Work on designing a strategy for the company you want to work for.  Do a tremendous amount of homework you can do from the outside – read analyst reports, get stealth demos, do market and customer interviews.  You don’t have to explain what you’re up to in terms of identifying the company.  You can say you’re interviewing for a CEO role in the sector.  Or even that you’re doing market research.  But proactively sending the hiring board a strategy deck and asking for the next meeting is a good way of differentiating yourself as a candidate and potentially accelerating a process.
  
• As a CRO candidate, go try to sell the company’s product.  Do it to a couple “friendlies” (e.g, people who are friends of yours, not active customers or prospects of the company you’re interviewing with) so you don’t tread on the actual business.  But create your own deck.  Get meetings.  Write up your experience.  Sending the CEO or board an email that says “Hey, I have a prospect already in the final stage of the funnel for you, can we work together to close her?” is a sure way to differentiate yourself as a candidate and potentially accelerate a process.

There may be a macro answer here as well.  The market is still choppy, and boards and CEOs are more conservative in most sectors and subsectors than they are in go-go times.  So that may be slowing things down in general and may even make it harder to act as-if.  But that doesn’t mean you can’t try.

I’ve been uncharacteristically quiet since April (I still hate non-competes and while I respect the right of the Chamber of Commerce to sue the FTC, I hope common sense prevails). Between then and now, we switched things up at Bolster, and my co-founder Cathy Hawley is now the CEO. Things are great there, and if you need any executive search help (Director to C-level or Board/Advisory/Fractional), let me know.

I’ve been hard at work on a passion project while I’ve been between things professionally, and I’m excited this week to announce the launch of my new podcast mini-series, Country Over Self: Defining Moments in American History. That link is to the web site where you can see the whole plan for the series.

Whether or not you’re a US History nerd like me, I hope you enjoy the Country Over Self podcast, especially since what I do is basically take a CEO lens to the whole subject.

Here are the links to the show on the three major podcast platforms and YouTube if you want a video option:

• Apple
• Spotify
• Amazon
• Youtube

I am taking a very nonpartisan approach to analyzing critical moments in American history to tell some of our shared stories and highlight some of our shared values as a country to play some small part in bringing us back together as a nation. This is NOT a political podcast, but it IS at least in part a response to this divisive election season and the environment the past 10-20 years, partly the product of a lifelong obsession with the American Presidency. Somehow, and I don’t know how this is possible, I’ve never blogged about it, but Brad has. My bibliography has grown a lot since then, but this is a good start.

My trailer (Episide 0, about 3 minutes long) is live as well as E1, on LBJ, which I just dropped today, all on all those platform show links above. I’ll drop 1-2 episodes a week until the end of the year when I’ll wrap up the series. I am so lucky to have been able interview the historians I have to produce this.

I am closing in some new CEO opportunties, so I’ll be back with more once those shape up.