It is all about humans... [Horizon 01 - Issue #10]

March 31 - April 6, 2025

We started this newsletter because we felt AI is hyped for the wrong reasons. All the other newsletters were either solely about tools or promoting automation at work with AI. Meanwhile, everyone at work freaked out that AI would take over their job.

We never thought about AI as a tool or an automation engine.

Sure, in the last 3-5 years, we have seen many new tools branded as “AI-driven” or with some parts of AI in them. Most are frontends for generic AI models (there is nothing wrong with that!). And yes, AI is amazing for automation, but that is nothing new. We have been automating our work for a long time with non-AI tools anyway. We still have a lot of work left to be done.

We always thought about AI as an augmentation tool for humans.

Maybe it is because we are marketers. We value human stuff like creativity, emotions, intuition, and storytelling. Maybe it is because we truly understand that a fully automated world is sad to live in. Where is the human element in there?

The more we learn about AI, the more confidently we can say the same thing over and over. AI is here to arm humans with extreme potential, put steroids on their creativity, and give back their focus and time to do meaningful work. AI is augmenting us. It is part of our workflow. It is part of our team. It is a team member, like you and me.

Companies that treat AI like a team member will 100X their workforce’s productivity. Those companies using AI as a tool will see some spike, of course, but not nearly as much as others.

In this edition of Horizon01, we share some examples of how others treat AI as part of their team. It’s amazing to see that the future is unfolding in front of us.

Lastly, this is our 10th edition at Horizon01. What a journey, even within this very short period! We hope you are enjoying our newsletter and find it valuable.

Thank you for your support and attention!

Peter & Torsten

A quick word from our sponsor, Superhuman AI:

Start learning AI in 2025

Keeping up with AI is hard – we get it!

That’s why over 1M professionals read Superhuman AI to stay ahead.

• Get daily AI news, tools, and tutorials

• Learn new AI skills you can use at work in 3 mins a day

• Become 10X more productive

Sign up and start mastering AI

Spotify: GenAI ads hit the airwaves

Spotify rolled out AI-powered audio ad creation tools within its Ads Manager (SAM) for advertisers in the U.S. and Canada. Marketers can input keywords, choose a voice style, and instantly generate a complete audio ad – script, voiceover, and even background music – using generative AI. The feature (dubbed “GenAI ads”) is in beta and comes at no extra cost to advertisers. Spotify’s decades of AI in music recommendation are now being applied to advertising, aiming to make ad production as seamless as creating a playlist.

Why it matters

A polished audio commercial traditionally requires a studio, voice talent, and budget. Now, even a small marketing team can produce professional-sounding ads on the fly. This could be a game-changer for brands with limited resources – local businesses or niche e-commerce sites – to get on air with quality ads without agency help. Spotify reports that 4 in 10 advertisers making audio ads on its platform have already tried the AI tool, signaling strong interest in automating ad copy and voice work.

Marketers can also rapidly iterate audio ad concepts – for instance, test two different scripts or voices for the same promotion – and deploy the best one. Got a seasonal campaign or flash sale? In minutes, you can have an upbeat voiceover with royalty-free music ready to stream to Spotify’s 500+ million users. It’s especially handy for brands targeting Gen Z on Spotify; the company notes that audio is an “always-on” channel for this demographic.

The bottom line is that audio advertising is expected to become faster, cheaper, and more personalized. The true test will be whether AI-made ads can actually grab listeners’ attention or just add to the noise.

Shopify as an organization is betting big on AI

Shopify CEO, Tobias Lutke shared his internal memo with the world in an X article. The memo describes how Shopify handles and works with AI internally and what future it is betting on. We think this is not just about Shopify - it can be a baseline for most companies in the following years.

Here are the best takeaways from the memo:

1. A new approach to hiring is to “Hire an AI before you hire a human.”

2. More on hiring: AI is a baseline expectation, and hiring managers will filter for AI-fluent candidates.

3. AI is a teammate, not a tool.

4. Prompting is a core skill, not a bonus. Top performers will be top prompters.

5. AI usage is measured similarly to any other business metric. This is huge!

6. AI-first prototyping is the new standard, which means shipping new product features should be expected 10X faster.

7. Since AI is a teammate, org charts will be blurred as they now include bots, not just bodies.

8. AI literacy is the new coding literacy. Everyone must be able to prompt, contextualize, and evaluate AI output.

9. The best part: AI is treated as a core layer in the tech stack. It is not a plugin, add-on, or bonus tool. It sits high as frontend, backend, and design. It is infrastructure.

What do you think? Are they too optimistic? Too bullish? We don’t think so.

Actively’s sales “Superintelligence”

Actively AI, a New York startup, just raised $22.5 million to build what it calls a sales “superintelligence” platform. The founders argue that typical AI sales reps (the kind that blast out automated cold emails) have “failed” because they focus on sheer volume rather than quality leads. Actively’s approach is different: they create custom AI models for clients that analyze the company’s CRM and data to pinpoint the most valuable prospects – prioritizing leads like a top human salesperson would. Under the hood, they lean on the latest “reasoning” AI techniques (forcing the AI to break down logic and double-check itself) to make more nuanced decisions rather than just spamming more emails.

Let’s talk about how NOT to use AI

Actively’s pivot might work out well, but we can learn a lot from why their first concept failed.

You see, more automation isn’t always better. Intelligent automation – that understands quality over quantity – is where AI needs to head in sales. You don’t need AI to be blasting out 10,000 emails. You need an analyst who tells you, “Here are the 50 best accounts to focus on this quarter.”

Suppose you can feed your sales data (past deals, customer profiles, engagement info) into a model. In that case, it will surface surprisingly insightful answers to questions like “Which leads in our CRM are most likely to convert right now?” or “Which existing customers have upsell potential this month?” In theory, this helps teams prioritize efforts and personalize pitches – focusing on leads that matter and even guiding on what messaging might resonate based on the AI’s analysis of account info.

But in the short term, don’t expect a plug-and-play miracle. Actively provides a custom model per client, which likely requires a good amount of your data and tuning. We’ll see in the coming months if this “sales superintelligence” lives up to its bold name or becomes just another AI spin on an old sales tactic.

Midjourney v7 – Evolution over revolution

Midjourney v7 has been launched, marking the first major update to the popular AI image generator in nearly a year. It introduces some new features: a voice-prompting mode (you can literally talk to generate images) and a faster “draft” mode that produces images 10× quicker at half the cost (albeit at lower quality, with easy upscaling after). Midjourney v7 also requires a one-time personalization step – rating images to tailor the AI to your taste – before you begin, aiming to deliver results more aligned with individual style preferences. The model architecture is reportedly “totally different” under the hood, with improvements in rendering fine details like hands, text, and textures, at least according to Midjourney’s CEO.

V7’s enhancements (better coherence, voice input, etc.) promise more convenience – imagine brainstorming an ad concept by speaking your idea and getting instant visuals. However, early user feedback is mixed. Many designers say v7 feels like only a slight upgrade (“more like v6.2” as one user quipped ) rather than a game-changer. Expectations were sky-high thanks to recent competition (OpenAI’s viral image model in ChatGPT set a new bar ), so some find the quality jump in v7 underwhelming. In fact, some prompts that looked great in v6 reportedly yield worse results in v7, and common pain points (awkward hands, jumbled text in images) are still not fully solved.

The marketer’s take

For creative directors and content marketers, Midjourney v7 is still a powerful tool to quickly visualize concepts – from storyboards to website graphics – but don’t expect miracles. The draft mode can speed up your workflow when you need many ideas fast, and personalization might gradually improve how well the AI understands your brand aesthetic. That said, you might want to keep Midjourney v6 handy for certain tasks if you prefer its output; v7 is an opt-in alpha, and you can switch back if needed.

Tip: Use voice mode for brainstorming sessions – it’s a novel way to spark creativity, even if it’s a bit buggy at times (some reported having to retry due to errors).

Overall, Midjourney v7 shows promise but isn’t the radical leap some hoped for – an important reminder that not every AI update will be a jaw-dropper.

Meta’s Llama 4 – Weekend surprise (and drama)

Meta unexpectedly launched Llama 4 – its latest open(ish) large language model – on a Saturday. This new family of models comes in three flavors: Llama 4 Scout and Llama 4 Maverick are out now (both ~17B active parameters with a mixture-of-experts architecture), while a giant Llama 4 “Behemoth” model (2 trillion parameters total, 288B active) is still training. They are multimodal, trained not just on text but images and video, giving them “broad visual understanding” out of the box. Perhaps most notably, Meta has tuned Llama 4 to be more open to answering controversial questions.

Why it matters

For the AI industry, Llama 4’s release is a big deal—it’s Meta’s answer to GPT-4-class models, and it’s (mostly) open-weight, meaning companies and researchers can get under the hood. Access to a powerful model without relying on OpenAI or paying usage fees is huge for businesses. We’re already seeing a thriving open-source ecosystem around Llama models, and Llama 4 will accelerate that, potentially spawning new specialized chatbots and creative AI tools built on top.

The timing of this release is also telling: Meta dropped it unannounced on a weekend, apparently spurred by competitive pressure (Chinese labs like DeepSeek were reportedly outperforming Llama 3, lighting a fire at Meta to catch up). It’s fair to say the rollout was a bit chaotic – even Meta’s marketing felt rushed. Maybe Meta’s marketing team took the weekend off, and the engineers hit “Deploy” anyway?

Controversy also struck immediately. Some experts accused Meta of benchmark gaming. Meta’s blog touted Llama 4 Maverick’s impressive performance on a chatbot arena test, but it turns out they used a special distilled version not publicly available for that score.

Implications for you

If you’re a developer or tech lead in a marketing/sales org, Llama 4 models (Scout for long documents, Maverick for chat/creative tasks) could be worth experimenting with for building in-house AI assistants or analyzing large datasets – Scout’s 10 million token context window is insane (it can digest whole databases in one go).

For less techy professionals, the influence will be indirect but significant: as Llama 4 gets integrated into products, you may soon find your customer support chatbot or your creative brainstorming tool got a quiet upgrade in brainpower. As a marketing or creative pro, expect a flood of new AI tools built on models like Llama 4 in the coming months, possibly with more customization and lower costs than the big closed APIs. And if you’re evaluating AI vendors, ask what’s under the hood – some may soon tout “Powered by Llama 4” as a selling point for flexibility and transparency.

BONUS: 20 insights from the Perplexity Fireside Chat with Roy Bahat (Bloomberg Beta)

Both of us are part of Perplexity’s Business Fellowship Program. During this program, we can network with other AI experts and attend amazing fireside chats with those pushing the AI changes forward in business.

We regularly share these insights with our subscribers. Last week, we had the opportunity to listen to Roy Bahat from Bloomberg Beta on building AI that's better for your employees.

Here are the top takeaways from this chat:

1. AI is “technology on steroids”—more powerful and faster, but still in a trial-and-error phase.

2. Every company should think like a tech company or risk missing out on AI’s advantages.

3. Individual use of AI tools is now more impactful than top-down initiatives. Those organizations will succeed that bet on AI-literate individuals rather than big-time organizational changes for AI.

4. AI can eliminate the most annoying and low-value parts of your day. Focus will be reclaimed with AI on things that really drive the business forward.

5. The best way to start with AI is by asking, “What’s the least valuable thing I do daily?” And then figure out how AI can do it for you.

6. Think of AI as an “always-on intern”—available for tedious, repetitive tasks. These tasks are safe to handle by AI, more-or-less error-free, and can horde up immense ROI for any business.

7. Don’t expect magic. Learning to use AI is like learning to ride a bike. It is part of practice and intuition. But once it “clicks,” you can’t forget it.

8. AI is best applied through aggressive trial and error on personal workflows.

9. Use AI to normalize inconsistent inputs (e.g., investor updates, pitch decks). AI is a great polishing engine, removing overlooked small errors and refining things to a baseline perfection.

10. Change starts at the individual level—“change yourself before changing the organization.” This 100X applies to businesses willing to bet on AI—start with the people, not with the organizational chart.

11. Corporate transformation is hard. Break initiatives into manageable, practical experiments. AI champions can drive these small, measurable steps, which AI newbies can replicate.

12. Familiarity with tools creates intuition—a key skill for future decision-makers. It goes hand in hand with creativity, which is needed to master a core skill: prompting AI.

13. Roy’s metaphor: Looms (automation), Slide Rules (productivity), Cranes (new capabilities). First, automate small tasks and get rid of the slack with AI. Then, human productivity can be enhanced by augmenting human work with AI so they can do better work faster. Together, these changes lead to new capabilities that wouldn’t have been possible without AI.

14. The most profound uses of AI create new possibilities—not just efficiency gains. This is why AI holds great promise for human potential.

15. AI will create jobs driven by human demand, not just technological supply. Of course, it starts with disruption - but the end route is clear and looks great.

16. “Artisanal” work (handmade, personal) could rise in value—think therapy or storytelling. Some things need a personal touch or just that “one more thing” that only humans can add (regardless of whether GenAI can replicate that).

17. Dignity in work comes from being needed by others, not task complexity. As humans, we are and will be needed even in an AI-driven world.

18. A $1B solopreneur company is not only possible—it may already exist. The most amazing outcome of AI is that we see one-man-shop businesses thrive better with AI than 100+ headcount companies without AI.

19. Many great businesses start as side projects. AI accelerates this potential. The best thing you can do is try to test AI daily. Play is back at work.

20. When pitching, don’t sell—show why you are the right person to win this race—every VC bets on the jockey, not the horse.

The next event will be with Amjad Masad, CEO of Replit. We’ll do a round-up of takeaways on that too.

Quick notes of the week

* Sponsored

• We upgraded our Beehiiv account, allowing us to launch a much better web experience for our readers. Please look at our new site and let us know what you think.

• Speaking of Beehiiv, if you want to launch your newsletter, use our link and get a 30-day trial + 20% OFF for 3 months.*

• We use a lot of AI notetakers. From Otter to Copilot, we tried a lot. It is the most obvious use case for AI for anyone flooded with daily meetings. Fathom makes taking notes, transcribing them, and sharing them with your team easier. Best feature: Fathom integrates with almost all meeting tools. Give it a try here and unlock 3 months free.*