The Office That Thinks + The Next !nvention

May 12, 2026

Today Roam, the Office That Thinks, launched our new website with an instant interactive product tour right on the home page! Forever, I’d dreamed of being able to give our site visitors a demo right on our site. This led to tons of experiments, none of which worked. We made Demo Roam, which was an incredible multiplayer demo, but was just too clunky, used actual AV that required permissions, and could take up to 10 minutes. The new Ro.am site achieves the right balance of instant interaction on the home page while showcasing the rich functionality of our Virtual Office Platform. This only was possible because Joe, our designer, was able to use AI Coding to get what was in his head onto a webpage directly. Joe designed and coded the interactive part with Claude, and then Jeff integrated into our existing infrastructure with Codex.

Check out the instant product tour on our site.

I’d also like to drop a breadcrumb about our next major !nvention. We’ve been inspired by the wild response we’ve seen to Claude Code and Codex on the Roam map. Our !nventors have been imagining what it looks like to run AI Agents inside your Virtual Office. Today’s orchestration is siloed, not collaborative and async. We see it differently: agents working out in the open, teammates jumping into a session together to push the next turn, agents summoned live to do work in a meeting.

That’s the Office That Thinks.

Onward,

Howard & The Roam Team

What Apple’s New CEO means for Virtual Office Platforms

April 21, 2026

I reflected a bit about the appointment of John Ternus as CEO, and I think that Apple is entering a new phase that may redefine not just its products, but the very concept of work itself. As the company transitions from the era of Tim Cook to a hardware-driven leadership model, it is increasingly positioned to champion a world where the “office” is no longer a place, but a distributed system powered by a mesh of intelligent devices. In that framing, Apple is not just building tools for productivity, it is laying the groundwork for fully realized virtual offices, where computation, communication, and collaboration happen seamlessly across devices, anywhere.

Ternus represents a return to Apple’s product-first DNA. His reputation inside the company is that of a meticulous builder, someone who prioritizes user experience over technological spectacle. At the same time, Apple has elevated Johny Srouji to oversee all hardware engineering. Srouji is widely regarded as the architect of Apple Silicon, one of the company’s most successful strategic bets in recent history.

I think that this alignment is critical because the future of AI at Apple will not be defined purely by models: it will be defined by where those models run.

We’ve been seeing Apple preparing for on-device AI for years. Neural Engines, efficient memory architectures, and tightly integrated OS’s have laid the groundwork for running increasingly sophisticated models locally. The rationale is straightforward: privacy, latency and reliability.

Apple already runs non-trivial machine learning workloads on-device, from speech recognition to image processing. In iOS 26, we got access to optimized Foundation Models that can now execute directly on mobile hardware, enabling a new class of applications.

For more complex tasks, Apple appears to be adopting a hybrid strategy, forming partnerships with external model providers, including systems associated with Google’s Gemini ecosystem, while maintaining strict control over privacy layers and orchestration.

By the time iOS 27 arrives, my expectation is not simply a “smarter Siri,” but a system-level coordinator: understanding user intents across apps, executing multi-step workflows automatically, and surfacing relevant information proactively

Apple has already laid much of the infrastructure for this through frameworks like App Intents and Shortcuts. The missing piece is an adaptive system capable of interpreting context and routing actions accordingly.

We, developers, remain first-class participants, exposing capabilities that the system can orchestrate through these frameworks. AI becomes a layer of composition and orchestration rather than a replacement for apps. OpenClaw but on your iPhone and not leaking your data everywhere in some sense!

So we’re gearing toward a more productive iOS. While not being itself a paragon of remote work, Apple effectively owns the stack of modern mobility: iPhone for communication, iPad for flexibility, and MacBook for high-performance portable computing.

What has been missing is deeper contextual awareness: systems that understand what users are doing, not just where they are. On-device AI enables complex workflows between expert apps, spanning multiple devices and platforms, while preserving the users’ data privacy. We’re preparing for these at Roam, with deeper system integration.

In this model, the office is no longer a location. It becomes a distributed system of devices and interactions, stitched together by intelligence.

If successful, Apple will not just build smarter devices. It will make intelligence ambient, available anywhere, at any time, without friction or compromise. And in doing so, it may finally deliver on a long-standing promise: that anyone, anywhere, can create, work, and think differently. It’s an exciting time to be developing and innovating on these platforms!

Don’t Build Your Company’s Brain Inside Slack. You Will Pay For It.

April 20, 2026

I spent years as CIO of a publicly traded company before I started Roam. In that time I watched a specific pattern play out again and again: a team adopts Slack for chat, and over the next few years quietly migrates many of its decisions, customer insights, and pieces of tribal knowledge into its channels and DMs. Not everything. But enough to produce the feeling that our company is running on Slack.

That is the moment you should be panicking. Because two things are true about the place you just put much of your company’s brain, and both of them are going to cost you: it only contains a fraction of how your company actually thinks, and the vendor that owns it is about to find out exactly how much you’ll pay to keep it.

Chat Is One Slice Of How Work Happens. Not All Of It.

The counter-argument, which I’ve heard from every Slack evangelist over the last two years, is that AI is going to fix all of this. Give it another six months and your Slack archive becomes a searchable, summarizable corporate memory. I actually agree AI will be extraordinary at organizing the chat transcript. That’s not the problem.

The problem is that the chat transcript is a small slice of how a company actually thinks, and it happens to be the least valuable slice. Decisions get made in meetings. Customer context gets created on sales calls. Strategy gets reframed in all-hands. Architecture gets sketched on whiteboards. Tradeoffs get explained in screen recordings. Slack sees the text residue of these events, if anyone bothers to write any of it down — which they usually don’t, because they just had the meeting.

So when you point an AI at your Slack archive and ask it what your company thinks about something, you get back a fluent summary of your company’s group chat. Not your company’s thinking. And that is worse than useless — it dresses up a fragmentary picture in the confident prose of a complete one. You get answers that sound authoritative and are missing the context that actually mattered.

Slack didn’t do this to you. You did this to yourself, because Slack was the easiest place to type something, and you slowly convinced yourself that what gets typed is what gets thought. That’s what Salesforce is counting on.

Then The Bill Comes Due

Slack has been owned by Salesforce since 2020. Salesforce did not pay $27 billion for a chat app out of sentimentality. They paid it because once a tool becomes the place your company thinks, you cannot leave. And when you cannot leave, the price goes up.

This is not a prediction. It is Salesforce’s playbook, run on every product they’ve ever owned: land cheap, expand into every workflow, wait until switching costs are unbearable, then move the pricing up and the features you need into higher tiers. Slack has been quietly walking this path for years — per-seat increases, AI gated to Business+, message history limits, admin controls parceled across tiers. The more valuable Slack has become to your company, the less leverage you have to push back. Salesforce’s sales team knows the number of channels you have, the integrations you’ve built, and the size of your archive. That is the number your renewal will be priced against. Not a penny less.

The Real Cost Is What You Can’t Measure

Buyers love to calculate the per-seat price. The actual cost of putting your company’s brain inside Slack is harder to spreadsheet. It is the cost of every AI-generated answer that sounds right and is missing the meeting where the opposite decision was made. Every new hire who onboards from a chat archive and comes out confidently wrong. Every strategic debate re-litigated because the version that got written down wasn’t the version that got decided. And eventually, a renewal conversation where the vendor knows you have nowhere to go.

A Different Approach

I didn’t just complain about this. I built differently.

At Roam, we built a virtual office — a single place where your team meets, presents, records, and chats. The decisions, the context, the recordings, and the text residue all end up in the same place because they all happened in the same place. That is what makes an AI layer actually worth something: when your assistant can see the meeting where the decision was made and the chat thread that followed, it can answer how your company actually thinks, not just how it types.

One plan, one price, every feature included, and we only bill for members who actually log in. The details are boring on purpose. The point is the philosophy: your company’s brain is too important to assemble from a dozen vendors’ scraps, and too important to rent from any one vendor whose business model depends on you being unable to leave.

The Bottom Line

Use chat for chat. Don’t let it become your wiki, your decision log, and your institutional memory all at once. And if you are going to put your company’s brain somewhere, put it somewhere that isn’t owned by a vendor whose entire corporate DNA is built around squeezing you the moment you depend on them.

Because Salesforce is very good at what it does. And what it does, eventually, is present you with the bill.

Who Wants a Headless Brain?

April 20, 2026

Last week Salesforce announced Salesforce Headless 360 for Saleforce, Agentforce & Slack.

The responses were mixed. Many jumped in offering their support, saying it was smart, others were trying to figure out exactly what it means. I personally do believe it is “correct”. Marc is brilliant at recognizing where the world is going and reshifting the positioning of Salesforce to capture the latest narrative. They’ve been on-demand, cloud, social, mobile, agent, and now headless.

Interestingly, I believe a headless CMS makes complete sense for a CRM system. Agents are going to need a sales system-of-record upon which to operate, and they will be best able to do so unencumbered by AI. Agents will analyze opportunities, update pipeless, suggest next steps, and assist a seller in the whole enterprise sales workflow. They’ll run CS.

However, a headless chat system - Slack - is a bit more challenging to digest. The point of a chat system is for human-to-human collaboration. Yes, Agents can be in the workflow when humans are in the loop. But going through a messaging platform is not an efficient way for Agents to talk directly. They will build their own APIs (and even possibly invent their own language!) to just get things done directly. Why go through an enterprise messaging system to get work done?

The argument for headless Slack is that Slack, with its messaging history, should be the context layer for the company. This is backwards. Chat is one data source among many. Meeting transcripts, code, CRM, office presence, project management, docs, email, calendar — most of what a company knows lives outside of Slack. The idea that you dump all of that into Slack so Slack can feed context to an agent gets the architecture exactly inverted. Context layers consume from systems of record. They aren’t themselves systems of record.

And, you have to just imagine how much Salesforce intends to charge and upsell for all of this. Salesforce is a notoriously ruthless pricing optimizer. Slack used to offer monthly billing by active users. No more. Now, it’s all prepaid upfront for the year. There are aggressive upsells and long term contracts. In the era of agents, don’t most companies want to move towards shorter-term usage based contracts?

So what is the head? Ben Lang of Cursor said it best:

An AI-native communication and work environment, not retrofitted. It pulls context from every system of record the company runs on. It’s where humans and agents coordinate, where work gets orchestrated and made visible, and where the interface is designed for both at once. Agents talk to each other through structured protocols, not enterprise chat. Humans work through the head, not alongside it.

Companies of the future want an AI native way to communicate and deliver context to AI agents, with AI native packing and pricing models. Agents will need a way to communicate as well. And companies will want a way to orchestrate and visualize the work going on. The system won’t be headless. It will have a giant head and be filled with a huge brain.

Roam Turns Your Talking into Doing

April 17, 2026

I’m more of an ideas guy. Coding is a tool to bring ideas to life, and I’ve always been good at it, but the exciting part has always been the ideas themselves. When I was working on my PhD and considering an academic career, the thing that genuinely appealed to me about becoming a professor was having students to handle the legwork while I focused on the high-level thinking. Turns out I didn’t need to become a professor; with the state of the latest AI tools, I think I’ve finally found my grad student.

At Roam we have a role we call !nventors, essentially product managers and engineers rolled into one. A big piece of that is coming up with ideas (or “!deas” as we cheekily spell it), but you also have to be able to see them through to implementation. Going from “we should build this” to actually building it is where momentum really matters, and I’ve been tuning a small workflow in Roam that’s been really nice for that.

I’ve configured my office in Roam so that it auto-records Magic Minutes by default. Whenever I have a conversation there, I get a transcript, a summary, and the ability to prompt the meeting afterward. So when a technical discussion wraps up and we’ve sketched out a feature or a change we want to make, I don’t have to scramble to write everything down or translate the conversation into a task. I just ask Magic Minutes to write me a prompt for Claude to implement what we discussed. This moves my job more towards simply discussing features and software design, letting AI take care of many of the simple implementation tasks.

The model has pretty consistently formatted those prompts in a markdown code block, so I added a little copy-to-clipboard button to code blocks (throughout Roam) to make grabbing it frictionless. One click, then paste it into the Claude CLI.

What makes this actually useful is that Magic Minutes has the full context of the conversation, so the prompts it generates are way more detailed than anything I’d write from scratch. I’d usually be too lazy to spell out all the nuance. Magic Minutes does it for me.

And if there are open decisions that still need to be resolved before the task is well-defined, I can work through those in the conversation with Magic Minutes first. It has the whole meeting context to pull from, so it’s not starting cold.

This is part of a bigger thing we’ve been thinking about at Roam: post-meeting workflows. Meetings generate a lot of follow-up, and most of it is either forgotten or handled inefficiently. We’ve been working on distilling the key workflows where AI can actually help automate the natural next steps. This is a simple one, but for engineers it’s a pretty direct path from technical discussion to implementation. Expect more here, both for tighter vibe coding integrations, and for some pretty different applications we’re working on.

Claude and Codex on the Map

April 15, 2026

We just shipped something at Roam that shows when your teammates are actively using Claude Code or Codex, live, on the office map.

Here’s how it works and how it got built.

The problem

Lots of us at Roam have individually been using Claude Code and Codex for a while, but we realized that we didn’t have a great sense of who else was using it or how much. The only signals were indirect. Someone mentioning it in standup. A PR landing with suspiciously articulate commit messages with lots of em dashes (or, less favorably, with a hundred tiny commits that each declared certain success but seemed to be going around in circles).

There’s a specific feeling in a physical office when a new tool starts spreading through a team. You glance at someone’s screen, see something unfamiliar, ask about it. That’s how tools actually propagate. Before Roam, remote work mostly killed that, but we are always looking for ways to bring it back.

The architecture

Both Claude Code and Codex support hooks: small scripts that execute at points in the agent lifecycle. UserPromptSubmit fires at the start of a turn, the PostToolUse hook fires after each tool call, and Stop fires at the end of the turn. We use that to ping a tiny HTTP server that Roam spins up on a unix socket in the Electron main process. When the hook fires, Roam knows the agent is active and sends updates to the server

The full hook command for Claude Code:

cat > /dev/null; curl -sf --max-time 0.2 \
  -X POST --unix-socket '~/.roam/roam-local-api.sock' \
  "http://localhost/.../post-tool-use?pid=$PPID" \
  >/dev/null 2>&1 || true

The cat >/dev/null at the start throws away everything Claude Code sends us. We don’t read your code, we don’t read the AI’s responses. We discard the entire payload and just register that something happened. This is privacy-preserving and auditable by anyone who can read a shell command.

The curl timeout ensures that even if Roam takes time to come back (which it shouldn’t) the hook doesn’t run long enough to trigger annoying status messages in Claude or the Codex TUI. (Unfortunately, for now, there’s no way around that in the Codex GUI, but the messages each have a fun hook icon next to them, so our Codex GUI users declare this not to be a blocker).

Where it got interesting: PPID vs session_id

For Claude Code, $PPID — the parent process ID — works well for session tracking. Each terminal session is a different process, so concurrent sessions get different IDs naturally.

Codex is different. When you have multiple threads running in the Codex UI, they can share the same parent process. So using $PPID meant stopping one session would clear the presence indicator for all of them.

The fix: Codex hook payloads include a session_id field. So for Codex we actually do read the payload, but only in the shell command and just long enough to extract that one field:

payload=$(cat)
sid=$(printf '%s' "$payload" | grep -oE '"session_id"..."' | head
-1 | sed ...)
unset payload

The visual

When Claude Code is active for someone, their office on the map gets an orange halo. Codex is blue. Running both? The colors blend via alpha compositing on a canvas element with a conic gradient rotating at 60 degrees/second, pulsing on a sine wave. Three-layer rendering: a wide 16px-blur halo at 80% opacity, a 4px-blur main body, a sharp core edge. Activity stops, it fades over 800ms.

As with other animations in Roam, we degrade to something much simpler when you’re on battery, on older hardware, if you explicitly choose a lower performance mode, or are in a meeting.

What’s missing

People have already asked for token counts on the map: see how much your teammates are burning and win tokenmaxxing contests. Turns out Claude Code’s hook system doesn’t expose token counts in any of its 24 event types. The data exists in the API response and transcript files, but hooks are deliberately lightweight. But we have some ideas.

The setup

It’s opt-in. When you arrive in your office, Roam checks if you have Claude Code or Codex installed (it just looks for ~/.claude/ or ~/.codex/), then shows a modal asking if you want to turn the feature on. You can enable or disable it any time in settings.

A hopeful note

I’m also glad to report that Codex was very gracious about being asked to add support for itself knowing that support for Claude was already in the code. Likewise, even after noticing that it was no longer the only coding agent supported by Roam, Claude remained upbeat when asked to make further changes to the feature. It even kept its snark to a 4/10 when confronted with Codex’s nitpicky but completely accurate reviews.

Designing a Breathing AI Indicator for Roam’s Virtual Office

April 15, 2026

We added a soft, living glow to Roam’s virtual office map so you can instantly tell when someone is vibing with Claude or Codex. The map is core to the Roam experience — you can glance and see who’s around, who’s in a meeting, who’s listening to music, or who’s deep in a DND session and doesn’t want to be interrupted. No scheduled meetings (for me at least), just natural drop-ins. As everyone started using AI more and more, we wanted a nice visual cue that felt alive but never got in the way.

My setup is pretty simple: I run Claude in Ghostty terminal right above my browser, with a quick Vite + React project underneath. It lets me tweak the code and instantly see the glow come to life on the map — super fast iteration, prompting away until I get the effect that I like. The glow sits around each room card like a soft halo around the edges. We used the real brand colors — Claude’s warm orange and Codex’s saturated blue — and turned them into a rotating gradient that slowly sweeps around the card, giving it that gentle pulsing effect. It fades in and out smoothly, and the more intense the AI work feels, the brighter and livelier the glow gets.

To make it look rich and glowy instead of flat, we layered three versions of the same glow: a soft wide outer haze, a medium body, and a crisp inner edge. Everything stays neatly clipped to the shape of the room so it never spills into the middle. When someone’s using both Claude and Codex at the same time, we just stack the two glows on top of each other — they naturally mix into a pretty purple overlap without any extra tricks.

I barely touched Figma at the start because iterating live in the browser felt way better for this one. Usually we go back and forth a lot with Claude and the Figma MCP for product work, and some features even get prototyped with Claude and Paper. But for this glow, being able to see it move and breathe right on the actual map made tweaking so much more intuitive. Once it clicked, we handed it over to engineering, added little Claude and Codex combo badges in the corner, and fine-tuned the animations. In the end, it makes the whole map feel a little more alive. It’s pretty cool watching AI happen in real time across the team — something that is unique to a Roam Virtual Office.

The AI-Native P&L

March 19, 2026

Much has been written about the death of SaaS. I believe SaaS won’t die, it will evolve. The dinosaurs will become extinct, but survivors will either be founded as or evolve into AI-Native birds. Just like dino-DNA is different from a bird’s DNA, the P&L of an AI-native SaaS company is going to take on an entirely different shape than its extinct cousins.

I realized this while building the go-forward 4 year financial model for Roam and I am stunned with how much we are going to accomplish for so little compared to what I did before at Yext. Here’s my convictions on the “AI-Native P&L”.

“The cost of software is going to zero, and the cost of acquisition is going to ∞”

-Shutterstock Founder Jon Oringer

R&D. As the cost of software goes to zero, R&D will stay flat over time. You may have a few engineers. Maybe they are 10x engineers. With AI they just became 1000x engineers. What this practically means is that R&D cost will essentially be fixed within a company’s lifecycle. You don’t need to hire more engineers as you grow your revenue.

Traditionally, companies have had to make the tradeoff between maintaining existing features and building new ones. This tradeoff no longer exists. AI Native companies can do more of both, for the same cost.

The latest SaaS benchmarking shows an average R&D spend of 24% at IPO. This will be much lower - 5-10% depending on how much you want to build.

R&D Bottom Line: 24% → 5%

Sales and Marketing. Salesforce spent 37% of its Revenue on sales and marketing, Hubspot 49%. The average $100m Saas Company spends 33%. It’s getting easier to make things. This leads to crowded markets. Traditional paid channels become expensive. I predict this goes up. Relationship-led sales can’t be replaced by AI Agents. And, companies will have extra money saved from R&D leaving extra budget to invest in growth. There will be a premium on customer acquisition.

S&M Bottom Line: 33% → 40%

G&A and Headcount. There’s been a lot of talk of the single person unicorn. I think this will happen. But most companies will need people, just way, way less people. I think Roam will get to $100m of ARR on just around 50 people, many entry level. That’s 1/10th of the number of people we needed at Yext to hit the same number. Most will report directly to me. At least I’ll know every person and what they’re supposed to be doing. This is an astonishingly low number. No middle management, less lines of random specialized ops roles, no “customer success”, less legal, less HR people. The fewer people you have, the fewer people you need to support them.

Also, there will be no office space expense. Roam spends exactly $0 on office space. At Yext we spent $20k/year on office space per person on leases alone, notwithstanding all the other stuff that came with it. No AI-Native company of the future is going to get locked down in long-term inflexible leases. They will hire sparingly, for the best cost and highest quality, and build in virtual offices. A physical office is not even AI-native.

The average Saas company G&A is 14% at IPO.

G&A Bottom Line: 14% → 7%.

Cost of Goods. 80% was always the dream margin, the average SaaS pubco is about 72%. Infra will remain the same, but cost of goods also includes customer service heads. AI will replace a huge part of this.

However, I believe companies will build so much AI as a feature into their products that the customer support savings will end up being a wash as spend goes to the foundation models.

Gross Margin Bottom Line: 75% → 75%

Revenue We’ve thus far discussed the expense side, let’s look at the revenue side for a moment. In a crowded market, I think companies with lower, simple pricing will have an advantage. So Revenue will be a bit harder to come by. I also think companies with large surface area have a big advantage over those with a narrow focus. I wouldn’t want to be the CEO of a Premium priced niche SaaS company with a traditional heavy P&L right now.

Also, long term multi-year contracts will be much harder to come by. Why would a big company commit in an era with rapid change?

This may not impact revenue, but it will drive down upfront cash collection and deferred billings.

Concluding Thoughts The SaaSosouruses will become extinct, but the evolved survivors will thrive as birds in their new lightweight form. They will be smaller, more nimble, and arm blooded. And, they have the opportunity to be more profitable.

AI-Native companies may not get as big as they did before, but there will be way more of them, and they will be more profitable.

It’s easier than ever to build. Just do it this new way so you build a modern car (faster and cheaper), not the old fashioned horse and buggy (slower and more expensive).

Bottom Line: +10-15% net margins realized

The Quest for Product Market Fit

March 10, 2026

More than any other question, founders ask me about product market fit. How do you find it? How do you know if you have it?

This is a good set of questions because the only job that matters for a founder is to find and hold on to product market fit. Every other job is a tangential side quest. Product market fit is like the sun. It’s the source of all life and movement for a company.

I’ve found product market fit 8 times in my life. There is no formula to discover it anymore than there is a formula to write a hit song. There is not even a standard measurement for it. What follows below are just some observations on the art of the quest itself.

• PMF is not binary. You can certainly NOT have it, but just like different stars have different strengths and sizes, so does PMF.
• Size: You can have PMF in a very small market. If your PMF is strong, you can have a monopoly in this small market. In college, Tom, Sean, and I made a site called “Just a Tip” that let users send anonymous tips telling friends about their annoying problems. It virally exploded overnight with millions of users. We had very strong PMF… in the “anonymous tips by email” market. Not exactly a huge market, particularly in our ability to monetize it.
• Strength: The strength of your PMF is directly proportionate to your organic growth. If leads or users show up out of nowhere without you having to lift a finger, that’s strong PMF. Just a Tip had very strong product market fit, just not in a very large or lucrative market.
• A Market of 1: If you build a product for yourself, you have at least satisfied a market of one. Presumably, there are other people like you. I made Confide because I wanted to use it myself. We made Roam because we wanted a Virtual Office for ourselves. Now, these are networked/team products, but if you make something you use and live in every day, you are living on the edge and more likely to succeed because you can tailor it to fit exactly what you’re doing.
• Change: Just as a star ultimately burns out over time, so does PMF. Early adopters dry up, channels lose their effectiveness, markets change. Competitors jump in. Most often you’re hit not by a direct competitor, but by a new paradigm that changes the demand profile for the product. Several times in my life I’ve had really strong market fit and slowly fallen out of it over time. A founder needs to be watching all of this on the edges and stay on top of it.
• Change can be GOOD: When Yext launched in 2006, the iPhone didn’t exist, and Google Maps was in its infancy. The explosion of these new mobile mapping services was an incredible market force that propelled the entire listings category into real-time updates on mobile devices. Yext was positioned perfectly to lift off from this with our PowerListings franchise in 2010. MapQuest printouts became Google Maps on the iPhone.
• Believe What People Do, Not What They Say: Everyone will lie to your face and tell you your product is great. If you offer it to them for free, they will give you positive feedback with a few meaningless nits to show they’re paying attention. It’s a waste of time. It’s misleading signal unless you make them pay. I endeavor to get harsh signals from the market, not happy signals from friends. There is nothing harsher than the market.
• Focus on the Customer, Not the Technology: Steve Jobs famously said you need to start from the customer and work backwards. One of the worst debacles of my career was Xone, a beacon-based product that failed at Yext. I was obsessed with a new low-energy Bluetooth capability that Apple built into iOS and was convinced it was the future of in-store experiences. I made the huge mistake of trying to figure out applications for the technology instead of just trying to consider what someone actually wanted.
• Bottom Up, Not Top Down: Going super granular is the most important thing I do. Correspondingly, the biggest mistake I see others make is starting at the market level and deciding to “enter the CRM space” with “AI” or “Cloud” or whatever. That’s just pattern matching to what’s already worked in the past and applying it to a new domain. That might work for investors. It may even work sometimes for other entrepreneurs! But it’s never worked for me. I’ve found you have to come up with original ideas, and that only comes from deep work, living down in the trenches of whatever you’re trying to do. The idea for Yext PowerListings came when I was running Yext Calls’ call tracking services and realized you could swap phone numbers in a MapQuest listing, and that businesses would want to control it.
• Services Companies Never Evolve into Product Companies: I can’t tell you how many times in my life I’ve met with a services company who said they were planning to do some service over and over again, build internal products to streamline those services, and then bring those products to market. I met with them later, and it never happened. I’ve never seen it happen once. If you want to make a services company, do it! If you want to make a product, do it!

Above all, I’ve found that you have to strike the right balance between dogmatically believing in the big-picture thing you want to do, while being willing to change the details constantly to figure out what actually works. As we founded the Yext Calls era business, I spent basically all my time with a small team running media experiments for brands like GymTicket.com or TVRepairman.com. GymJungle.com was a Google-like search for gyms. We tried everything. Very, very few things worked at all. But when something did, we grabbed it and ran with it.

One last point. It’s way, way, way harder than you think. It takes way longer. It involves way more pain. Look at every single major tech company from Amazon to Google to Meta to any of Elon’s companies. They all have an arc of failure every few years where the world is convinced they will not survive. They look nothing like they did in 2005. They’ve required constant effort and reinvention led by the founders (look what’s happened to Google since Brin returned to reshock the company).

Yet decades later, they beat on, because the quest for product market fit is an unending founder epic that starts all over again the second you think you’re done.

RoboHoward Menu: 2000 cals, 225g protein

March 5, 2026

I ate the same thing every day and got jacked.

Through trial and error, I believe I have perfected the perfect daily menu. This RoboHoward Menu is designed for me to:

• Stay Lean
• Maintain and slightly build muscle mass
• Be fully energized without ever feeling hungry or craving. It tastes great and I eat it almost every day.

The menu is 2000 calories and 225 grams of protein. I discovered a secret item that holds it all together for me. Here’s the full menu:

I’ll now go through each item, starting with the chicken. A eat a whole pack of white meat chicken strips. It’s portioned in a bag like this. 250 calories and 50 grams of protein. This is the foundation of the menu because the ratio of protein to calories is so high. Here’s the kind I get, it’s from Whole Foods, but any store has this:

The next staple is this wood roasted salmon. You can get it from Trader Joe’s or Whole Foods. It has 240 calories and 24 grams of protein. The protein density is lower on Salmon, but it has Omega-3 and healthy fats. But above all it tastes amazing. The peppered one tastes like pastrami.

I heat the chicken and salmon, maybe add some butter lettuce and then I add mustard. It’s a super first meal of the day at 2:00pm. (I don’t think the timing really matters much, this timing and order is simply what works for me given my routine).

Meal #2

At 7pm, I have 93/7 ground beef, a bag of broccoli and a roasted Japanese sweet potato. The ground beef must be 93/7 - that is what keeps it lean. Most steaks and other things have much higher fat and calorie content. I don’t add any oil or butter to cook. Just salt and pepper, or the kindred spice mix. The ground beef has 680 calories and 96 grams of protein. It tastes like a hamburger!

I mix the broccoli in with the beef. The broccoli is key because it’s a huge volume of food and has lots of minerals and Vitamin C but comes with very little calories.

Then, the secret menu item which was a big breakthrough: the japanese sweet potato. For me this was a huge breakthrough because I do not really like normal sweet potatoes. They’re too watery and soft. I prefer the starchy texture of a regular potato. This is precisely what the Japanese sweet potato has, but it’s sweeter and better. I am increasingly convinced it’s a perfect food. You roast it with a bit of salt and a touch of oil, it comes out insanely great. It brings the whole thing together and totally satiates me.

Mix everything up and you get this:

Meal #3: Dairy Snack

You can eat this in the morning or at night. I do it at night. The baseline is 2 fat free greek yogurts and I mix in a scoop of vanilla or chocolate whey. I slice a banana with it. Optionally, I add in some berries. Strawberries, sometimes dark frozen cherries. It tastes like ice cream. I don’t know why more people don’t just eat this. It has 52 grams of protein!

I also have an apple I can slice up anytime and crunch when I feel like it.

Results

The RoboHoward Daily Menu delivered real results. Starting on Jan 1, I ate it each day for 57 days in a row. I got a DexaScan on Jan 1 and another on Feb 27. The results were stunning. I lost 8.4 pounds of fat and put on 1.5 lbs of muscle. Also, I had 0.0 … yes, zero visceral fat. I’ve never been leaner and I’ve never felt better.

Conclusions

Eating the same thing every day may seem extreme. Maybe it is for a lot of people. But I’ve never had any problem acting independently and just doing my own thing. I had to basically opt-out of many things that normal human people do, like enjoying restaurants.

I don’t recommend everyone do this. But I’m sharing it because if it worked for me, maybe it will work for someone else out there. I won’t stay on this 365 days a year, but I will do it 5 or 6 days a week. I have more energy than ever, I’m more locked in than ever in building my company, and I feel amazing each day. I hope reading this will cause someone to take an action that brings you the same incredible feeling of positivity that I feel as I am writing it.

Onward!

Howard