AIHOY AInauten,

Welcome to the latest edition of your favorite newsletter.

Quickly, two Anthropic notes before we leave you alone with it: by the time you read this, Fable 5 should be available again. And Claude Sonnet 5 is here, landing almost at Opus level for knowledge work.

What that means for you: for most everyday work with Claude Cowork and similar setups, Sonnet is now enough. We tested it, it works really well, and if you are paying only around 20 euros a month, this should let you do noticeably more with Claude.

Now let’s get into today’s topics. Lots of practical things to try, a free skill on top, and one bigger industry question at the end.

Here is what we have for you today:

Let’s go.

Not just another AI newsletter. MavSource aggregates updates from all major AI newsletters, podcasts, company news, AI labs, and hundreds of other sources — then summarizes what matters, analyzes emerging trends, and adds founder commentary. One 5-minute daily email. Free.

Subscribe For Free Now

This week, we are deep in AI video again.

So much is happening here right now that it is worth taking another look and testing a few things yourself.

Last week we went deep into Higgsfield. It is one of the most powerful video tools on the market because it is not just one model. It is a whole platform that gives you access to almost all relevant video models.

The team behind it is fast, agile, and creative. New features keep landing, and the point here is simple: this is worth testing if you do anything with creative production.

The fastest way to understand how powerful Higgsfield can be is one feature: the Higgsfield Supercomputer.

It is basically an ultimate creative agent. You give it a simple prompt, and it can develop the whole thing from A to Z.

You type a normal-language briefing, for example: “Create a campaign for product X.”

The Supercomputer plans the steps, chooses the right models, runs several subtasks in parallel, and delivers finished assets.

In our test, we asked it to create a 30-second UGC spot for a vitamin C serum, first generate three product reference images, create a UGC character named Sophie, and turn everything into a TikTok-style video with voiceover and subtitles.

The Supercomputer then created the assets: character, product, product shots, UGC script, and the video itself.

First attempt, impressive quality. Almost too easy.

That shows what Higgsfield can do. Product shoots, online-shop visuals, consistent AI avatars, UGC videos with the same character, all of that is possible there.

One last strong point: Higgsfield has an MCP and CLI, so you can connect it to Claude, Codex, and similar tools and let your agents create assets for you.

The best place to start is the Higgsfield Supercomputer. Create an account, choose a plan, and play with it in chat. Thanks to the new Sonnet 5 model, it is faster and cheaper too.

Important downsides:

Still, it is a very strong tool with a huge number of creative templates and powerful features.

Next week we will continue with more ways to use AI video. We are learning a lot ourselves right now and will share more soon.

Usage-based and hybrid pricing models are reshaping B2B revenue—but for finance teams, they're also introducing serious complexity around revenue recognition, forecasting, and operations.

Join Rebecca Schwartz, Co-founder of Tabs, and Amit Dhir, Partner at PwC, for a live session on June 10th from 1–2PM EDT. They'll break down how modern pricing decisions ripple through your financial workflows—and how to scale without adding manual overhead.

You'll leave with practical frameworks, real-world examples, and a clear path forward.

Register now to secure your spot—and if you can't make it live, a recording will be available.

Can't wait? Explore more Tabs webinars to catch up on past sessions.

Save My Spot

If you have been reading us for a while, you know we love using AI for strategy work and as a sparring partner.

We like taking concepts that used to require expensive coaches, trainers, or a full-day workshop and turning them into practical AI workflows.

Today’s setup is this: you have a plan, you ask your AI, “Is this good?”, and it finds ten reasons why it will be great.

You feel validated, start executing, and a few weeks later exactly the obvious thing blows up in your face.

That happens because language models tend to agree by default. They want to make you happy.

Today’s practical hack gets around that. We use an old proven technique: the premortem.

Instead of asking afterward, “What went wrong?”, you fast-forward: It is one year later. The plan failed. Tell me how it died.

That one sentence changes the job. AI stops polishing your plan and starts describing every path where it can fall apart: failure story, hidden assumption, early warning signs, and kill criteria.

This is not a toy method. It comes from Gary Klein and was published in Harvard Business Review. Daniel Kahneman called it one of his most valuable decision techniques.

We have used it for two strategic AInauten decisions, and it showed us blind spots we would otherwise have missed.

We prepared a full setup so you can use it immediately: our version is a bit sharper than the ones currently floating around, with a “what did we miss?” countercheck and clear kill criteria.

You can download it here:

→ Get the premortem for free

The download includes two versions: one as a skill for Claude Code and similar agents that builds a report file, and one you can paste into chatbots. Pick the one that fits your workflow.

These skills are especially useful if you already have a second brain, because the machine can include your real context.

Have fun with it, and send us your experiences.

Let’s end with a topic that many consultants, service providers, and agencies are wrestling with right now.

The old service-business deal was simple: smart people spend hours on hard problems, and the client pays for those hours.

For many, the last few months were golden: AI tools around Claude Cowork and Codex became useful, work got done much faster, but clients were still willing to pay old hour volumes.

That is changing.

The Wall Street Journal reports that large consultancies want to move away from hourly billing because AI makes a lot of work faster and cheaper.

No, they do not really want to. They have to. Clients know that work got easier, so they want it cheaper.

The billable hour punishes the fast. If you become more efficient, you can bill less value.

Large consultancies are already rebuilding: Deloitte has shown its consultants a curve where classic hourly consulting shrinks sharply by 2035. At McKinsey, more than a quarter of global fees are already tied to outcomes instead of hours.

The short-term path for service providers is probably away from hourly billing and toward fixed-price packages with clear goals and outcomes.

In some cases, 100% performance-based deals can also make sense. In our performance marketing campaigns, for example, we have long worked on CPL and CPA models: clients pay per lead or sale.

That puts us at risk. It works in online marketing, but of course not everywhere.

As always, it comes down to incentives.

The fact is: the window of great AI arbitrage is closing. Consultants, service providers, and agencies have to adapt.

We think that is positive. If everyone is incentivized to make projects successful, that is healthier. It also gets more complex.

If service providers take on more risk, margins can rise significantly too.

That is it for today. As always, we hope something in here helps you move forward.

See you next week.

Reto & Fabian from AInauten