There are three common promises of new AI startups:

1. Replace staff with agents
2. Free up staff for strategic work (a repackaging of #1 but targeting the middle manager budget)
3. Give consumers a faux talent upgrade (vibe coding, writing, art, cheating on exams, etc.)

All of these are missing the real value: LLMs, at their best, increase tempo.

What is tempo? Tempo is not speed, though speed is often a result. Tempo is the rate of the process that produces speed.

With a higher tempo, you learn faster, you make faster decisions, and you correct mistakes faster.

Consider the shift from physical to digital photography. When I was growing up, camera film required time to develop: you would need to finish a roll of film, and then take it to a store to get it developed.

For any worthwhile photos, you would need an expert, both because you lacked the professional skill, and because getting it wrong would take too much time and money.

With digital photography, the roll of film is gone: any photographer can see their results immediately. This allows them to try 50 approaches instead of two, learning by doing.

Eventually, consumers would learn the same skills, whether on a DSLR or an iPhone. And now? Most teenage girls take better photos than I can.

AI is creating a similar shift across every business function, providing an opening to increase tempo, learning, and competency faster than before.

First, consider software product development. (I'll discuss sales and marketing later.)

Tempo in Software Development

Six months ago, I hacked together a chrome extension to scrape some complex webpages. At the time, vibe coding was still new and it mostly consisted of code completions and chat. I learned that I could theoretically run the web scraper using Playwright or Puppeteer, but I hit several dead ends as the context window was too small to understand my needs, and the training data likely had some old versions polluting the recommendations.

Fast forward to this week: I was able to pick the project back up on Monday, and have a refactored app running by Wednesday. With the expanded capabilities now available in Cursor, I added functional tests, optimized some of the algorithms, and learned how to run local processes.

What happened here?

• The tempo to pick up old code has increased 10x
• The tempo to learn principles has increased
• The tempo to ship product has increased

All of this, and I had leftover mental room to improve more things in the app, rather than running out of energy at the finish line.

Tempo in Customer Development

An emergent property of this software development tempo can actually be faster customer development (if you allow it.)

Around November, I identified a new opportunity: LLMs can rapidly lower the cost of migrating from legacy platforms, and there are a few $100B companies that are ripe for attack.

Most devs would see this opportunity and immediately start writing code, convinced that the product is the most important thing.

Instead, I noticed the abundance of the product, and the likely speed that I could build it, and focused instead on interviewing potential customers.

In two hours I had an enriched list of 1st degree connections who would be perfect fits for this new service. Convinced of its clear superiority, I contacted the first 20.

In four days I had the key insight: no one wanted this product.

What happened here?

• Finding potential customers was 10x faster
• Contacting potential customers was 10x faster - writing the personalized message was now trivial
• Getting to failure was 20-50x faster (one week instead of one year)
• The number of equivalent experiments is now 20-50x higher

Failing after a week feels a helluva lot better than failing after a year!

Tempo in Sales

If product and customer development are mostly single player, sales is multiplayer: the tempo of the prospect matters as much as the tempo of the salesperson.

Still, there are some incredible opportunities here to improve sales tempo:

• Follow ups with meaningful collateral can be produced 10x faster
• Preparing for a meeting is 10x faster when an LLM can scan all of the context
• Writing proposals is 10x faster
• Sales coaching post-call is 10x faster

The biggest gain in sales, though, is the ability to scale Founder-led Sales.

Salespeople reduce tempo, because they lack agency.

Prospect requests a feature? The salesperson either promises it (without knowing it can be done) or checks with the product team to see what is possible.

A 30 second question can become a four week delay. Meanwhile, the prospect is evaluating alternatives.

Prospect requests a feature from the Founder? Its a yes or a no.

That's tempo.

Founder-led Sales has limits: there are only so many meetings you can prep for and attend, so many proposals you can write, so many follow ups. But LLMs accelerate all of these.

Tempo in B2B Marketing

Most of B2B Marketing is simply small products made for small audiences, by people who understand neither how to make products nor how to listen to their audience.

So the typical Whitepaper, once a staple of deep tech startups positioning themselves for the future, is now a 3-7 page PDF that can be produced by Perplexity in a few hours.

The typical case study can be hallucinated into your template in a similar amount of time.

Blog posts? Newsletters? All can be generated 10x faster.

Generic marketing is now 10x faster. But let me tell you what this means for great marketing.

Most companies won't try great marketing until they've exhausted the generic marketing.

Unfortunately, most startups fail before they exhaust the generic marketing options.

LLMs accelerate this.

When you can implement every generic marketing idea 10x faster, you'll either find something that works, or you'll begin looking for meaningful marketing ideas.

In practice, this looks like more companies hiring and firing their CMOs 10x faster. Agencies will be on far shorter leashes.

Eventually they realize what the founders kind of already knew: marketing only works when you listen to your audience and reduce friction for them. That is, if you can get your marketers to talk with customers.

And actually, LLMs can accelerate this too!

LLMs make it possible to create synthetic versions of your customers and prospect, so marketers can build for them.

This is more than buyer personas (though most implementations will be similarly shallow.)

First, LLMs make it possible to rapidly learn from every recent customer call, integrating each conversation into the total context about your market.

Second, LLMs can actually create synthetic customers: profiles of your audience that marketers can talk to, 24/7, to understand what they want.

Once marketers combine this level of customer context with the ability to produce small products 10x faster, they actually begin to become revenue generating members of the team.

The Compounding Effect

What makes tempo worthwhile is the way it compounds in ways that efficiencies cannot.

When your product development tempo increases 10x, your customer development tempo increases 10x, and your sales tempo increases 10x, the end result is not a 30x improvement: its closer to 1,000x.

A founder can use AI to build a product prototype, immediately get it in front of prospects, gather feedback, iterate on the product, and close early customers, in weeks instead of months or years.

Imagine two startups founded the same day with identical resources. One builds products with LLMs but otherwise runs a traditional customer development and sales process, the other optimizes for tempo in every function.

After six months, the traditional startup has built 10x more than average. Nice. Their product is highly refined for their stage. But they've only talked to 50 potential customers, iterated on their product a few times, and closed maybe five deals.

The compounded startup, meanwhile, has had 500+ prospect conversations, which fueled 20+ product iterations. They've likely closed 20-40 deals, but more importantly, they've taken their product, messaging, and sales approach through more feedback cycles. They are far more likely to have something like product market fit.

Not the sum of the parts

You might be reading this and nodding your head: "That's what we do."

Many founders have pushed every team to adopt AI into their work, and think this means they are compounding gains.

They're wrong.

The magic here isn't in writing a blog post 10x faster, shipping 10x more code, and holding 10x more meetings with prospects, all at once.

The magic here is in doing these together, within a tight feedback loop.

Jumping 10x faster on your right leg, and then jumping 10x faster on your left leg, is going to feel like quite a workout, but it doesn't get you nearly as far as running 10x faster.

The enduring advantage

Another reason tempo matters is how it compares to efficiency in distribution among players.

Everyone with a smartphone, today, has access to a camera superior to anything I saw as a child. Everyone has a more efficient camera, and the cost savings of not processing film, yet not everyone is a great photographer (I'm not.)

The difference between a great photographer and my mediocre photos comes down to the tempo and learning.

I never stop to look at my photos, reflect on how to improve them, and try again. In fact, the abundance of photography makes me feel like I don't need to improve.

Professionals (and teenage girls) take the time to learn from their photos and improve.

I expect the same will happen with LLMs.

Everyone has the same access to the efficiency of LLMs. Every student can put their homework assignment into an LLM and get their homework done for them. This is efficient.

The winners will use LLMs to more deeply understand the assignment, chase down their curiosity around the topic, write the assignment they want to write, and then get feedback from an LLM so they can improve their thinking.

In business, everyone has the same access to LLMs. Everyone can put in the same list of content topics and ask for 500 word blog posts to help them rank for SEO. Everyone can ask for a LinkedIn post about an adjacent topic to their market, and then hit publish with their eyes closed.

Everyone can vibe code their way to a passable MVP.

The difference between the winners and losers will be how much they learned by doing.

The content spammer who thinks "its just a numbers game" and publishes 10x more, without reflecting on what worked or didn't work, is like that person at the gym running 90 minutes on the treadmill every day who never seems to get faster.

The dev who builds their product "in only 60 hours" but still takes 6 months to ship, rather than 6 days, squanders all opportunity to keep up.

LLMs make us much better at our jobs, but what if our job is to become something better?