Human Sandwiches & AI Courtrooms: The New Rules of Work
- X —iO
- Jun 9
- 6 min read
We have officially solved the "individual productivity" problem, but in doing so, we accidentally created an organizational bottleneck. Right now, 88% of companies use AI, but only 6% are capturing enterprise-wide value from it.
Why? Because handing everyone an AI account doesn't redesign work. A useful agent needs context and reliable data. Without it, the AI isn't a colleague, it is just a very enthusiastic marten in your operating system, exhausting your team with endless, average output and chewed cables.
The smartest companies at Accelerate Tomorrow are ruthlessly redesigning how work actually happens. Here are their rules.
1. Make It "Crappy" to Build Confidence
When trying to get a non-technical workforce to adopt AI, most companies offer a boring coding course and wait for perfection. But as Cornelius Frey, CEO of NucleusLinks, pointed out, perfectionism kills innovation. The real enemy is psychological fear. To get teams building, force them to be playful. Have them build a geography quiz for their kids or a walking vacation tour, and make them demo it to the team in two hours, -no matter how "crappy and buggy" it is.
2. Build a "Courtroom" for the Big Decisions
Messy speed is great for learning, but disastrous for existential business choices. Human leaders usually decide with their emotions first and justify it with logic later.
Because AI is probabilistic, it will often just feed us what we want to hear. To combat this bias, Alexander Pöllmann (Viessmann) and Helmut Scherer (Iteratec) built an "AI Courtroom". They set up a "Red Buffet" agent (an extreme optimist finding upsides) to debate a "Blue Buffet" agent (a ruthless pessimist finding risks).
It sounds exactly like me pitching brilliant business ideas to my partner, only to be told to "refine your prompt, D." Suddenly, ruthless criticism looks like a massive competitive advantage.
3. Capture Your "Tribal Knowledge"
An AI trained on the internet is brilliant, but it has no idea how your specific business runs. Without context, it hallucinates generic fluff. To fix this, Florian Kunzke, Global Director AI Strategy at SAP, explained how tools like SAP Signavio capture "company memory"—the tacit knowledge hidden in emails and Slack messages.
Until you connect AI to your messy, internal data, it cannot make decisions. Take the internal rule: "coconuts must only be sourced from Brazil" . I learned this the hard way in Rio—the coconuts are exceptional, but pour cachaça in it and try dancing samba, most of us would short-circuit like a broken robot. Your AI needs that same local context.
4. Eat the "Human Sandwich"
Are massive, highly-siloed marketing teams fired? Don’t panic yet. Volker Harbrecht from Meta explained that humans simply cannot manually create the hundreds of variations needed for algorithm personalization today. Instead, 100-person silos are being replaced by agile units of just five strategic "captains" using the "Human Sandwich" model:
Top (Human): Set brand strategy, budget, and guardrails.
Middle (AI): Generate massive scale and targeting.
Bottom (Human): Analyze, course-correct, and optimize.
Never let an AI dictate your brand’s soul. Keep the strategy human and let the AI handle the exhausting scale.
5. Beware the Jevons Paradox
If you automate customer service, you save money and eliminate jobs, right? Breath! Not exactly. When Klarna automated 66% of its support, they laid off 700 workers—but shortly after, had to rehire them to maintain service levels.
Matthias Goehler from Zendesk calls this the Jevons Paradox: when you make a service cheaper and easier to consume, people consume much more of it. By making 24/7 support instantly accessible, your interaction volume will likely spike. Automate the basic tasks, but retrain your human employees to handle complex, proactive VIP relationships.
6. Build Proactive Sales Engines & Lego Blocks
This shift from reactive to proactive isn't just for customer support. Benedikt Nolte (Plato), alongside Timo Burkat & Murat Yücel (Meesenburg), showed how B2B wholesale teams, who can't mentally manage massive product catalogs, are using AI to uncover hidden revenue. By digesting data across 8 billion product-customer combinations, AI identifies logical cross-sell links, turning reactive order-takers into proactive advisors.
To scale this across the enterprise, Allianz Tech recommends building reusable AI "Lego blocks" rather than isolated tools. For example, they built a fully autonomous system in Australia using seven specialized AI agents (triage, fraud, data, payout, resolution…) working together to process food spoilage claims, leaving only the final review to a human.
7. Abandon the "Pilot Island” and Build a Real Wedge
Right now, companies are handing out chat accounts and hoping for the best. But as Johannes Foertsch from OpenAI noted, basic activity is not a business advantage. You need a "wedge" redesigning one high-value workflow deeply before expanding
Julian Eckerle (Vercel) and Florian Kunzke (SAP) warn against forcing AI to follow rigid, hard-coded steps. Instead, Leo Reuter (Vercel) suggests giving the AI a "virtual file system" packed with your company's context so it can explore data on its own. When you pull AI off its isolated "pilot island" and plug it directly into your core enterprise data, autonomous software starts making the decisions and sending the requests itself.
8. From Dead Keyboards to the "Flow Era"
For decades, digital efficiency meant clicking menus or typing commands into sterile little boxes. According to Gordian Braun from ElevenLabs, that "friction era" is officially closing.
We are entering the "flow era" of voice AI that understands tone and emotion in real-time. Klarna uses it to resolve issues 10x faster across 70 languages, and Deutsche Telekom uses it to diagnose Wi-Fi problems. The craziest part? It handles human impatience perfectly. During the demo, when an annoyed user interrupted the bot mid-sentence, it adapted instantly without breaking flow, -a brilliant feature for grumpy customers, and an absolute win for the humans who no longer have to deal with them.
9. AI for Humans, By Humans
Have you noticed how everyone on the internet suddenly sounds like a 19th-century philosopher crossed with a tech bro? As Gregor Schmalzried from Der KI-Podcast pointed out, AI loves to package completely banal ideas as profound. We must shift our mindset from asking AI to "do this for me" to "help me do this," keeping it a collaborative companion.
This lack of original thought isn't just a marketing problem; it goes all the way up to the C-suite. Burkhard von Spreckelsen, Chief Development Officer at Elmos Semiconductor delivered a reality check: AI is just a probabilistic model learning from the past, designed to feed us what we want to hear. If Steve Jobs relied strictly on historical Blackberry data, the iPhone would have launched with a plastic keyboard. Despite exponential hardware advancements, leaders must lean into distinctly human traits: purpose, character judgment, and critical thinking.
10. The Governance Trap: Think Like Auntie Miao
Many enterprises are stuck in a governance trap, requiring endless approvals just to use AI for a safe 10% efficiency gain on legacy systems. Meanwhile, AI-native operations in China are succeeding through a fearless "try it and screw it" mentality.
Take "Auntie Miao," for example. As insights from Qihua Wang revealed, this woman in her mid-50s built a ground-up AI-native business processing €3 million a year completely alone. She didn't wait for an enterprise IT roadmap. She simply uses AI to automatically translate and run her live streams globally.
Aside from skills, competencies, and experience, the cultural foundation (of how people from China, EU, and USA think) is very different. Comfort slows you down, but an existential crisis builds motivation, curiosity, and the willingness to take risks.
The Bottom Line: Stop Bolting AI onto Broken Processes
We are transitioning out of the era of the shiny chatbot and into the era of fundamental operational redesign. It is time to leave the Pilot Island and start building real Wedges powered by Proactive Advisors and reusable AI Lego Blocks. If you simply bolt AI onto an old, inefficient process, you are just making your old way of working the permanent ceiling.
The companies that will win the next decade are the ones leading with distinctly Human Traits. They are building Human Sandwiches to manage massive scale, evolving their human teams to handle VIP relationships, and setting up AI Courtrooms to challenge existential risks. They are jumping headfirst into the Flow Era asking the AI “help me do this” with the fearless, boundary-breaking mindset of Auntie Miao.
If you don't have millions on the line, stop waiting for the perfect IT roadmap. Dig up your Tribal Knowledge, build something a little Crappy today, and put the marten to work.
A massive thank you to Till Schmid (ATS Founder) and team for organizing ATS 2026. Thanks to the brilliant speakers who provided the frameworks, reality checks, and the courage we need to actually redesign the present:
Cornelius Frey (NucleusLinks), Alexander Pöllmann (Viessmann) and Helmut Scherer (Iteratec), Florian Kunzke (SAP), Volker Harbrecht (Meta), Matthias Goehler (Zendesk), Benedikt Nolte (Plato), alongside Timo Burkat and Murat Yücel (Meesenburg), Firas Ben Hassan (Allianz Tech), Johannes Foertsch (OpenAI), Julian Eckerle & Leo Reuter (Vercel), Gordian Braun (ElevenLabs), Gregor Schmalzried (Der KI-Podcast / ARD) and Burkhard von Spreckelsen (Elmos), Qihua Wang (Lagardere) Thanks Dr.Oliver Krause (Advantum) and Mark Turrell (unDavos) for the mentorship, Aditya Gupta for the tips, a big shout out to AinTheMachine DJ
More about #AI on X-iO and down the #rabbithole🕳️🐇
Comments