The Next Web TNW

AI agents are now being embedded across core business functions globally. Soon, these agents could be scheduling our lives, making key decisions, and negotiating deals on our behalf. The prospect is exciting and ambitious, but it also begs the question: who’s actually supervising them? Over half (51%) of companies have deployed AI agents, and Salesforce CEO Marc Benioff has targeted a billion agents by the end of the year. Despite their growing influence, verification testing is notably absent. These agents are being entrusted with critical responsibilities in sensitive sectors, such as banking and healthcare, without proper oversight. AI agents require clear programming, high-quality training, and real-time insights to efficiently and accurately carry out goal-oriented actions. However, not all agents will be created equal. Some agents may receive more advanced data and training, leading to an imbalance between bespoke, well-trained agents and mass-produced ones.  This could pose a systemic risk where more advanced agents manipulate and deceive less advanced agents. Over time, this divide between agents could create a gap in outcomes. Let’s say one agent has more experience in legal processes and uses that knowledge to exploit or outmanoeuvre another agent with less understanding. The deployment of AI agents by enterprises is inevitable, and so is the emergence of new power structures and manipulation risks. The underlying models will be the same for all users, but this possibility of divergence needs monitoring.  Unlike traditional software, AI agents operate in evolving, complex settings. Their adaptability makes them powerful, yet also more prone to unexpected and potentially catastrophic failures. TNW City Coworking space – Where your best work happens A workspace designed for growth, collaboration, and endless networking opportunities in the heart of tech. For instance, an AI agent might misdiagnose a critical condition in a child because it was trained mostly on data from adult patients. Or an AI agent chatbot could escalate a harmless customer complaint because it misinterprets sarcasm as aggression, slowly losing customers and revenue due to misinterpretation.  According to industry research, 80% of firms have disclosed that their AI agents have made “rogue” decisions. Alignment and safety issues are already evident in real-world examples, such as autonomous agents overstepping clear instructions and deleting important pieces of work.  Typically, when major human error occurs, the employee must deal with HR, may be suspended, and a formal investigation is carried out. With AI agents, those guardrails aren’t in place. We give them human-level access to sensitive materials without anything close to human-level oversight. So, are we advancing our systems through the use of AI agents, or are we surrendering agency before the proper protocols are in place?  The truth is, these agents may be quick to learn and adapt according to their respective environments, but they are not yet responsible adults. They haven’t experienced years and years of learning, trying and failing, and interacting with other businesspeople. They lack the maturity acquired from lived experience. Giving them autonomy with minimal checks is like handing the company keys to an intoxicated graduate. They are enthusiastic, intelligent, and malleable, but also erratic and in need of supervision.  And yet, what large enterprises are failing to recognise is that this is exactly what they are doing. AI agents are being “seamlessly” plugged into operations with little more than a demo and a disclaimer. No continuous and standardised testing. No clear exit strategy when something goes wrong.  What’s missing is a structured, multi-layered verification framework — one that regularly tests agent behaviour in simulations of real-world and high-stakes scenarios. As adoption accelerates, verification is becoming a prerequisite to ensure AI agents are fit for purpose.  Different levels of verification are required according to the sophistication of the agent. Simple knowledge extraction agents, or those trained to use tools like Excel or email, may not require the same rigour of testing as sophisticated agents that replicate a wide range of tasks humans perform. However, we need to have appropriate guardrails in place, especially in demanding environments where agents work in collaboration with both humans and other agents. When agents start making decisions at scale, the margin for error shrinks rapidly. If the AI agents we are letting control critical operations fail to be tested for integrity, accuracy, and safety, we risk enabling AI agents to wreak havoc on society. The consequences will be very real — and the cost of damage control could be staggering. source

As an American born and raised in New York City, I’ve seen the power of US entrepreneurialism to change the world. The ambition, ingenuity, and relentless drive that have powered the country’s economy for generations have also been a global force for prosperity, stability, and innovation. Yet now the US is retreating into an aggressive and unpredictable form of unilateral bullying. I am deeply concerned — not just for America, but for the world.  For the past few years, I’ve watched these developments from Europe. I’ve settled with my family in the Netherlands, where I work as CEO of cultivated leather startup Qorium. I’ve been impressed by the world-class infrastructure and public services, but I’ve also encountered frustrations for which Europe is famous: slow decision-making, risk aversion, and onerous regulation. Yet over time, I’ve come to see these as features to be worked with rather than bugs to be squashed. They are evidence of a system that values durability, collaboration, predictability, logic, and long-term thinking over speed, spectacle, and zero-sum “I win, you lose” politics. They offer Europe a unique advantage in the global race for technological leadership — and the continent can seize it with regulatory change. But its success hinges on a difficult shift: adapting its culture. On the regulatory side, the signs are positive. Europe is forging a new path that supports technological ambition with public trust, democratic legitimacy, and stability.  Take the AI Act. Often dismissed by Americans as slow and bureaucratic overreach, it is in fact the first serious attempt anywhere in the world to create a harmonised framework for the development and deployment of AI. Rather than leaving developers in a regulatory grey zone or overwhelming them with patchwork national laws, the act establishes clear risk categories and compliance pathways. Yes, it demands responsibility — I’d argue too much right now — but it also offers certainty. In sectors like biotech, healthtech, and critical infrastructure – where uncertainty is often a greater deterrent than regulation – this is crucial, especially as America becomes increasingly erratic. Consider also the Digital Services Act and Digital Markets Act. These regulations don’t just attempt to rein in Big Tech excesses; they lay the groundwork for a more competitive, open digital ecosystem. Combined with GDPR, now a de facto global standard (albeit not without its flaws), these frameworks show that Europe is no longer content to be a rule-taker in the digital age. It is becoming a rule-maker, and increasingly, the place where responsible innovation can get done. TNW City Coworking space – Where your best work happens A workspace designed for growth, collaboration, and endless networking opportunities in the heart of tech. This regulatory clarity is already making a difference. European universities and research centres are seeing rising applications from non-EU nationals. International PhD and postdoctoral researchers, particularly in ethically sensitive or publicly impactful fields, are beginning to choose Europe not just as a stopover but as a base. Venture capital is responding too, with notable upticks in funding for deep tech startups across Germany, France, and the Netherlands. Europe’s approach may not generate the overnight paper unicorns of Silicon Valley, but it fosters sustainable, scalable innovation with real-world impact. On the cultural side, however, there is work to do. Process, structure, and legislation, no matter how effective, cannot replace the passion, optimism, and relentless drive that underpins innovation in US entrepreneurship. Europe needs to learn to believe in itself, and if not to “move fast and break things,” at least move faster than it does now. Frankly, it needs to learn to work harder and more — a mindset that’s not easy to acquire. Yet overall, the progress is positive. Pan-European initiatives – from Horizon Europe to the European Innovation Council – are addressing these gaps, with billions in coordinated funding and support for high-impact research and tech transfer. Perhaps most encouragingly, there is a growing sense of urgency among European policymakers that innovation isn’t just about competitiveness – it’s about values, focus, and prioritisation. This contrasts starkly with the mood in the US. Higher education is under siege, with books banned, entire departments defunded, and educators fired for teaching history factually. Federal rhetoric is openly hostile to basic scientific facts. Research funding has been weaponised. If the US ceases to be a safe haven for open inquiry and intellectual freedom, the best and brightest minds will go elsewhere. And they already are. A growing number of international students are choosing Canada, Australia, and EU countries over the US, citing visa challenges, political instability, and cultural hostility. American researchers, too, are beginning to take up posts abroad, often for the same reasons. The long-term effects of this brain drain will be profound. Europe, meanwhile, is sending the opposite message: that science and innovation are public goods, that truth is not a partisan issue, and that education is a right, not a privilege. For international talent – whether you’re an AI ethicist, a quantum physicist, or a biotech founder – that message is magnetic. Let’s be clear: Europe is not perfect, and I still believe in the power of American innovation. But the global competition for talent and innovation is accelerating. The rules are changing, and Europe is playing the long game – with a strategy rooted in values, clarity, and collaboration. As someone who grew up believing America was the place where the future got built, I now find myself looking across the Atlantic and thinking: the future can be built here too. Europe can thrive as a stable, open, truthful hub for innovation — a zone of free inquiry between America’s instability and China’s ideology.  If Europe maintains its foundations while embracing a pro-business, pro-innovation culture that rewards risk, hard work, and dynamism, it has a once-in-a-generation opportunity — not just to compete, but to lead. The world desperately needs it. source

Vibe coding is redefining who can build software. By enabling code generation through natural language prompts, it’s quickly gained traction among startups and indie developers. But the biggest opportunity lies ahead: the enterprise.   The rapid rise of Lovable — which recently raised a $200mn Series A at a $1.8bn valuation — illustrates the remarkable progress of vibe coding. Having backed the Swedish startup at the seed stage, I see this as just the beginning. What’s next? A fundamental upheaval of who can build software — a cultural shift set to transform entire industries.     The vibe coding revolution The disruptive power of vibe coding is already evident. More would-be founders can develop software today, which means a bigger potential talent pool and new catalysts for innovation. Creative entrepreneurs with amazing product ideas will no longer be stalled by a lack of coding expertise or access to developers. The significance of this cultural shift shouldn’t be overlooked. If technical strength is no longer enough to stand out, the spotlight shifts to user experience. User-centric design and community-building will emerge as the primary measures of success. Consumers stand to benefit most, as the vibe coding era ushers in companies that emphasise exceptional experiences and intuitive interactions. TNW City Coworking space – Where your best work happens A workspace designed for growth, collaboration, and endless networking opportunities in the heart of tech. The appeal of vibe coding to tech founders and indie developers — exemplified by Lovable’s rapid growth — is understandable. It aligns perfectly with the “move fast and break things” mentality that favours speed, aesthetics, and agility over processes.  Enterprises could also reap the benefits. By using vibe coding to empower non-technical employees, they could produce new apps for internal and external use cases. This could ease the cost and resource constraints of today’s development status quo, which depends on in-demand technical skills and talent.  But “could” is the key word here. Vibe coding today clashes with the realities of enterprises, shaped by regulation, risk aversion, and stringent demands for security, compliance, and auditability.  It’s no secret that current vibe coding projects create a lot of technical debt. This friction means vibe coding apps aren’t quite ready for enterprise use. However, the vibe coding startups that adapt to enterprise demands stand to capture a significant market opportunity. The enterprise opportunity To crack the enterprise market, vibe coding platforms and products will need to show comprehensive audit trails that can explain everything built by the AI. This transition is already becoming a reality. Github CoPilot, for example, strikes a great balance between autonomous code suggestions and edits with checks and balances, creating a professional-grade AI coding product for mainstream industries. Vibe coding startups can replicate this for enterprise users who lack the coding knowledge of GitHub users. This presents a big opportunity for first movers like Lovable, as well as emerging players who can tailor verticalised solutions to enterprise needs. Once the risk-based concerns are addressed, companies can realise the potential of slashing software build costs to near zero. Better products will flood into the enterprise market. They’ll either be built with vibe coding, offer enterprise users ways to apply vibe coding in their roles, or provide a combination of both. The improvements to user experience and product quality threaten legacy enterprise software providers, which have long relied on cumbersome and inflexible solutions. Vibe coding holds the potential to disrupt this stagnant model, fostering a more productive and innovative enterprise software landscape. From an investor perspective, vibe coding startups remain highly compelling due to their disruptive potential. Investors will closely monitor startups’ ability to retain users and transition beyond initial pilots to sustained, long-term adoption — a crucial indicator of success. Demonstrating sustainable growth post-launch — especially by moving into the enterprise mainstream — will be key to building further investor confidence. The next phase of vibe coding Vibe coding has upended traditional software development, showcasing promising results in startup ecosystems. By making it easier to build great products, it can elevate digital user experiences across countless use cases. The next test for vibe coding startups is breaking into the enterprise market. There’s every reason to think that first-movers like Lovable, as well as new players, can rise to that challenge — and take the vibe coding revolution to much greater heights. The future is bright. source

Watching robots awkwardly flop around, cause robot body pile-ups on the soccer field, and accidentally lose their heads while taking part in a 1500-metre sprint at the first Robot Humanoid Games in China was not only entertaining, it was a reminder of just how far robotics has come — and how far it still has to go. While humanoid robots still struggle to walk across a stage, in other corners of the world automation is quietly revolutionising industries. At Picnic Technologies, the Netherlands’ fastest growing online supermarket, robots are compiling your grocery orders so delivery ‘shoppers’ can get them from the warehouse to your refrigerator as fast as possible. It’s these innovations that have helped the once humble startup scale rapidly to compete with supermarket behemoths like Albert Heijn. The company’s CTO, Daniel Gebler, recently shared the secrets behind the company’s success with TNW founder, Boris Veldhuijzen van Zanten, as they drove through the streets of Amsterdam in the latest episode of “Kia’s Next Big Drive.” Check out the full interview — recorded en route to TNW2025 in Kia’s all-electric EV9 — by clicking on the image below: TNW City Coworking space – Where your best work happens A workspace designed for growth, collaboration, and endless networking opportunities in the heart of tech. Caption: Gebler and Veldhuijzen van Zanten winding through the canals on their way to TNW2025. But while Gebler holds a PhD in AI and is driving automation at scale, he’s clear that robots won’t replace humans entirely. Bananas and champagne Previously, Picnic’s ‘shoppers,’ who fill orders and deliver them to customers’ doors, had to walk around large warehouses picking out each item. Now the company’s fully-automated fulfilment centres in the Netherlands and Germany are helping to lighten the load (and the number of steps shoppers have to take) by automating the item picking process with robotic arms. At its newest order fulfilment centre in Oberhausen, Germany, Picnic is capable of processing up to 33,000 online orders per day, serving up to 200,000 households. The warehouse employs 1,500 robots… and 1,000 humans. Why? Because some tasks are still better handled by people. Bananas and champagne: Robots struggle with irregularly shaped items, fragile goods like eggs, or high-value products like champagne bottles. Packing efficiency: Humans can easily rearrange crates to maximize space, while robots require predefined layouts. They also have trouble opening boxes. Final touches: Even in highly automated centers, the last step — packing items into a customer’s delivery box — is still done by hand. To work around these limits, Picnic uses product whitelisting to decide which orders a robot can fill. For example, an order containing bags of crisps and heavy bottles of soda would be a no go for a robot. So, as robots evolve will they ever completely replace Picnic’s warehouse shoppers? “Absolutely not. As mentioned, it isn’t our goal to replace them either, but rather to use robots to boost our warehouse’s performance. Shoppers remain at the core of our warehouse operations, with robots complementing their efforts,” says Picnic software engineer Jhon Mauro Gomez. In other words: automation makes Picnic faster and more efficient, but it’s a collaboration, not a takeover. Could AI be coming for your boss? (Don’t get your hopes too high) The rise of AI is also transforming what “management” means inside companies. But Gebler believes AI won’t necessarily eliminate management entirely — it will reinvent it. “Most likely what we now have as management won’t exist anymore,” Gebler said. “The relevance of ownership — owning what you build, owning what you run — will become even more important. Because everybody will be a designer, a builder, and also an operator.” This shift gives teams more autonomy and room for experimentation. At Picnic, developers have used that freedom to: Launch return deliveries: Customers can now return retail items from other brands using Picnic’s delivery vans — making the fleet more efficient. Offer meals, not just items: Families benefit more from curated meal packages rather than piecing together individual products. The rise of “AI-free Fridays” Gebler is also pushing for “AI-free days” — dedicated time where developers ditch AI tools and sharpen their human skills. Because while AI can crunch data, it still can’t improvise like a human. Whether in grocery warehouses or corporate boardrooms, the future isn’t humans versus robots — it’s humans with robots. Automation is best at handling repetitive, structured tasks. Humans shine in areas requiring adaptability, creativity, and judgment. From bananas and champagne to AI-free Fridays, Picnic is proving that the future of work is not about replacement, but reinvention. Image credit: “BvOF RoboCup2013 – RoboCup Soccer Nao” by RoboCup2013 is licensed under CC BY 2.0. source

AI agents are now being embedded across core business functions globally. Soon, these agents could be scheduling our lives, making key decisions, and negotiating deals on our behalf. The prospect is exciting and ambitious, but it also begs the question: who’s actually supervising them? Over half (51%) of companies have deployed AI agents, and Salesforce CEO Marc Benioff has targeted a billion agents by the end of the year. Despite their growing influence, verification testing is notably absent. These agents are being entrusted with critical responsibilities in sensitive sectors, such as banking and healthcare, without proper oversight. AI agents require clear programming, high-quality training, and real-time insights to efficiently and accurately carry out goal-oriented actions. However, not all agents will be created equal. Some agents may receive more advanced data and training, leading to an imbalance between bespoke, well-trained agents and mass-produced ones.  This could pose a systemic risk where more advanced agents manipulate and deceive less advanced agents. Over time, this divide between agents could create a gap in outcomes. Let’s say one agent has more experience in legal processes and uses that knowledge to exploit or outmanoeuvre another agent with less understanding. The deployment of AI agents by enterprises is inevitable, and so is the emergence of new power structures and manipulation risks. The underlying models will be the same for all users, but this possibility of divergence needs monitoring.  Unlike traditional software, AI agents operate in evolving, complex settings. Their adaptability makes them powerful, yet also more prone to unexpected and potentially catastrophic failures. For instance, an AI agent might misdiagnose a critical condition in a child because it was trained mostly on data from adult patients. Or an AI agent chatbot could escalate a harmless customer complaint because it misinterprets sarcasm as aggression, slowly losing customers and revenue due to misinterpretation.  According to industry research, 80% of firms have disclosed that their AI agents have made “rogue” decisions. Alignment and safety issues are already evident in real-world examples, such as autonomous agents overstepping clear instructions and deleting important pieces of work.  Typically, when major human error occurs, the employee must deal with HR, may be suspended, and a formal investigation is carried out. With AI agents, those guardrails aren’t in place. We give them human-level access to sensitive materials without anything close to human-level oversight. So, are we advancing our systems through the use of AI agents, or are we surrendering agency before the proper protocols are in place?  The truth is, these agents may be quick to learn and adapt according to their respective environments, but they are not yet responsible adults. They haven’t experienced years and years of learning, trying and failing, and interacting with other businesspeople. They lack the maturity acquired from lived experience. Giving them autonomy with minimal checks is like handing the company keys to an intoxicated graduate. They are enthusiastic, intelligent, and malleable, but also erratic and in need of supervision.  And yet, what large enterprises are failing to recognise is that this is exactly what they are doing. AI agents are being “seamlessly” plugged into operations with little more than a demo and a disclaimer. No continuous and standardised testing. No clear exit strategy when something goes wrong.  What’s missing is a structured, multi-layered verification framework — one that regularly tests agent behaviour in simulations of real-world and high-stakes scenarios. As adoption accelerates, verification is becoming a prerequisite to ensure AI agents are fit for purpose.  Different levels of verification are required according to the sophistication of the agent. Simple knowledge extraction agents, or those trained to use tools like Excel or email, may not require the same rigour of testing as sophisticated agents that replicate a wide range of tasks humans perform. However, we need to have appropriate guardrails in place, especially in demanding environments where agents work in collaboration with both humans and other agents. When agents start making decisions at scale, the margin for error shrinks rapidly. If the AI agents we are letting control critical operations fail to be tested for integrity, accuracy, and safety, we risk enabling AI agents to wreak havoc on society. The consequences will be very real — and the cost of damage control could be staggering. source

Private equity deals hit an all-time high in 2021, peaking at a total value of more than $1tn, with an average deal size exceeding $1bn for the first time. Founders were media darlings, valuations soared, and investors raced to get a piece of the action.   By 2023, many of those same companies — such as Klarna and Stripe — had lost billions in value. Klarna’s valuation plummeted by 85% from its 2021 peak of $45.6bn to $6.7bn in 2022. Stripe also fell dramatically, from $95bn in 2021 to $50bn in 2023.   Fast forward to today, and even more tech companies are folding, from no-code platform Builder.ai to fintechs Frank and Stenn. Yet investors are still ploughing fortunes into risky ventures — particularly in AI. Case in point: the Thinking Machine Labs raised an eye-watering $2bn seed round without a single proven product.    In a race to invest in the latest, most eye-catching tech, generalist investors with money to spend are focusing on personalities and promises, failing to scrutinise product value, market fit, and opportunity.    TNW City Coworking space – Where your best work happens A workspace designed for growth, collaboration, and endless networking opportunities in the heart of tech. With $1.2tn in buyout dry powder still waiting to be invested — around a quarter of it idle for four years or more — pressure on dealmakers is intensifying. And when investors start chasing opportunities without much scrutiny, their behaviour starts to look like dating on impulse. It calls to mind a meme: the Hot Crazy Matrix. Lessons for private equity from an internet meme The Hot Crazy Matrix emerged in a viral YouTube video from the noughties. It offered a “scientific” framework for evaluating women based on two axes: “hot” and “crazy.” Problematic? Absolutely. But also, weirdly applicable to private equity.  This chart doubles as a startup investor guide. Credit: Tactical Response Crew / YouTube Now, before anyone gets HR involved, we’re not rating investors on physical attractiveness. In this version, the “hot,” horizontal axis represents specialism. The more niche your expertise, the further right you sit on the chart. Think of it as an investor who knows how to speak the language. Someone who gets the investment thesis right off the bat.  Then there’s the “crazy,” vertical axis. In our private equity version, it represents how big and bold a fund is. At the bottom: large, generic investors who skim the pitch deck and call it research. At the top: niche, smaller operators who actually understand what they’re buying and how to build value.  How to spot danger The left-hand side of the matrix — the no-go area — in this scenario doubles up as our danger zone. This is where we see large, generic funds with deep pockets throwing huge sums at investments without specific understanding or knowledge of the sector, product or commercial proposition. These guys jump in with limited ability to interrogate the details. This isn’t business — it’s gambling! A bit of fun while it lasts, but don’t be surprised if you end up losing your shirt.   To illustrate my point, see Exhibit A: Builder.ai.   Dazzled by the promise of a revolutionary new AI-powered platform, investors, including Microsoft and Qatar’s Sovereign Wealth Fund, poured more than $450mn into the business, pushing its valuation past $1bn. But beneath the glossy pitch, critical flaws went unnoticed: revenue figures had been overstated by 300%, and tasks marketed as AI-generated were actually being completed by a large team of human workers. The oversight was costly — and a stark reminder that deep pockets without deep understanding can lead to expensive missteps. The investors would sit on the left-hand side of our matrix.  On the other end of the scale, we’ve got the niche specialists. These PE investors know their stuff, but they’re often smaller. And while brains are great, a growing business needs brawn too — a firm that can actually move the needle.   Wife material In the sweet spot — or “marriage zone” — is a fund that’s big enough to commit to a full buyout, but also knowledgeable enough to unlock real value in its niche sector, A capital markets data company, for instance.  It’s a virtuous circle: expertise guides the investment, and the investment, in turn, builds even greater expertise.  But what about the mythical unicorn? Does a huge private equity firm with deep, specialist expertise exist? Maybe. But finding one is like dating someone who’s rich, kind, funny, and knows how to fix your Wi-Fi. Possible, but you might be waiting a while.  In private equity, as in dating, it pays to look beyond the surface. Flashy pitch decks and billion-dollar valuations might be tempting, but if you don’t know what you’re getting into, you might wake up next to a portfolio full of regrets. When capital is abundant, but clarity is scarce, private equity needs more than just enthusiasm — it needs discernment.   The Hot Crazy Matrix may be a pithy internet meme, but its reimagining offers a serious lesson: the smartest investors aren’t chasing the hottest trends — they’re bringing together deep expertise with commercial insight. Because in the end, just like marriages, the best deals aren’t the flashiest — they’re the ones that last.  source

A decade ago, startups often equated success with rapid headcount growth. The formula was simple: build a product, raise a round, hire fast. Bigger teams meant bigger bets. But the rulebook is getting rewritten as a new generation of startups scales with leaner teams and fewer people. They’re not building out sprawling customer support or sales teams, and seem to be automating what once warranted entire departments. Their growth is quite remarkable.Cursor, which became the fastest-growing SaaS company in history, generated $200mn in revenue with 30 employees. Midjourney made $200mnn with 40. Ben Lang’s site Tiny Teams tracks these small-but-mighty operators, with several emerging from Europe too. Sweden’s Lovable has a 25-strong team and achieved a $1.8bnn valuation just over six months after launching. Vlayer Labs, headquartered in Warsaw, secured $10mnn in pre-seed funding with 20 employees, while Berlin-based Juna AI raised $7.5mn with a seven-person team.These aren’t outliers. Startups of all kinds are slimming down, particularly in the consumer-facing and fintech sectors. In 2022, Carta, a platform that tracks startup equity and hiring trends, found that the average seed-stage consumer startup had 6.4 employees. By 2024, that number had dropped to 3.5. What’s next could redefine the startup world entirely. Imagine a startup that scales to millions in revenue without hiring a single employee. No head of growth. No support team. Just code, bots, and maybe a founder or two in the control room. In 2024 – a lifetime given AI’s breakneck speed – OpenAI CEO Sam Altman predicted the rise of one-person unicorns. As bleakly dystopian as it sounds, the zero-workforce startup is closer than we think. 1.0 burn multiples become the norm While AI and automation are undoubtedly accelerating tech’s “Great Slimdown” – call it the Ozempic era for startups – this is not the only driver. Since the heady days of the 2021 bull market, the VC industry has contracted, prompting companies and investors to commit to efficiency with renewed fervour. The number of active venture capital firms in Europe dropped by 30% between 2022 and 2024, while venture funding to European startups saw a steep fall in Q3 2024. It plunged to $10 bn, the lowest since Q3 2020, a 39% YoY decline and a 36% drop quarter-over-quarter. Tobias Bengtsdahl, a partner at VC firm Antler in the Nordics, has noticed the shift first-hand. “We invest in the very early days, when there’s only one to three founders, but already we’re sensing that they can go so much further and build so much more than in the past,” he explains. Prior to Antler, Bengtsdahl founded personalised video message platform Memmo in 2019, which hit $10mn in revenue within two years, and expanded to 150 employees in 20 months. “In the zero interest craziness of 2021, it was more important how many engineers we had than how prudently we were spending,” he says. “This is a very real pendulum swing, where investors are increasingly cautious and careful now.” European startups also face a significantly tougher debt market: bank loans now carry rates of 9-13% for early-stage firms, up from near-zero a few years ago. That caution is layered over a deeper, structural change. Founders aren’t just spending less, they’re building in a different way with AI and automation. Engineering teams, for example, are smaller due to code generation tools like GitHub Copilot and Tabnine, which enable developers to ship faster with less. In 2024, GitHub’s own research found that developers who used GitHub Copilot completed tasks 55% faster than the developers who didn’t. In the US, hiring for software development roles is down over 15% year-on-year, according to CompTIA. Customer support (often one of the first departments to scale) is increasingly handled by AI. GPT-powered assistants like Intercom’s FinAI Agent resolve up to 80% of Tier 1 support tickets instantly, according to internal benchmarks. It’s a similar story in sales and marketing, where tools like Jasper and Breeze, HubSpot’s AI suite help generate outreach, content, and campaigns in a fraction of the time. For many founders, this AI infrastructure isn’t a stopgap — it’s a hiring philosophy, with early-stage CEOs delaying key hires and in some cases, deciding against filling certain roles. As ultra-lean, AI-native startups become the norm, VCs are rethinking what constitutes a ‘scalable’ business. Molly Alter, partner at Northzone — the global VC firm behind Spotify, Klarna, and Trustpilot — says her team is spotting signs of product-market fit earlier, without the corresponding buildout. Expectations around revenue growth and burn multiples have morphed too. “A 1.0 burn multiple used to be incredibly impressive — a sign of disciplined, efficient growth — but it isn’t that rare anymore,” says Alter. “The bar has shifted.”It’s reshaping the day-to-day of venture scouting too. “Before, part of the job was scanning LinkedIn for headcount spikes, but that’s no longer a good signal,” says Alter, who focuses on vertical SaaS and AI investments. “We have to get on the phone directly with every founder to understand what’s really going on.” She’s often asking probing questions like: Why is the team lean? What’s being automated, and what still requires human judgement? Is there a plan for where humans will matter most, and where they won’t? And the once-assumed virtue of hiring itself might even be a red flag. “Investors and VCs in the market tell me they have no appetite for hyperscaling anymore,” says Roei Samuel, CEO and co-founder of networking platform Connectd. “At SXSW earlier this year, it was clear — unless you’re hiring for deep tech or there’s genuine defensibility around the expertise, they actually see hiring as a negative.” Dismantle, and then dismantle again AI is already reinventing roles at startups. Credit: Pixabay TNW City Coworking space – Where your best work happens A workspace designed for growth, collaboration, and endless networking opportunities in the heart of tech. Some founders believe the lean era isn’t about shrinking teams — it’s about rebuilding them from scratch around AI. “If a company simply sprinkles AI on top of their existing

Remember the movie Dodgeball? That ridiculous scene where the coach makes his team run across a busy highway? The logic: “If you can dodge traffic, you can dodge a ball.” Europe’s approach to AI feels similar: if you can survive our labyrinth of rules, you can survive anywhere.  Conversations with European companies about AI rarely begin with “What can it do?” Instead, they open with a sigh and ask, “Are we allowed to use this?”  For most industries, that’s a creativity-killer, but legal professionals thrive in regulatory swamps. Europe’s swamp is about to become its competitive moat. The paradox: red tape as rocket fuel Regulatory complexity around AI hasn’t slowed legal tech down. AI law tech startups attracted nearly $2.2bn in 2024 alone, accounting for around 79% of all funding for legal-related startups. The 💜 of EU tech The latest rumblings from the EU tech scene, a story from our wise ol’ founder Boris, and some questionable AI art. It’s free, every week, in your inbox. Sign up now! Prevailing wisdom says regulation strangles innovation. In European legal AI, it’s the opposite, partly because the industry is already marinated in compliance, and partly because no one outside Europe wants to deal with this mess. Love it or hate it, the General Data Protection Regulation (GDPR) has become the de facto blueprint for privacy legislation and shaped European laws, business practices, and digital trade norms since 2018. It’s influenced data privacy policies further afield, from Brazil’s LGPD and China’s PIPL to frameworks in Japan and India and even legislation in US states like California, Virginia, and Colorado. The more the EU sets global norms, the more legal AI systems built here will seem “export-ready.”  In this context, regulation becomes the product, as European lawyers sell their advice on the very rules everyone else dreads. If your AI tools can review contracts, undertake due diligence, or identify data protection risks under GDPR, they can do it anywhere. Legal professional standards, confidentiality and privilege are therefore protected by the red tape. Beyond LLMs The market is also learning. According to a 2025 Axiom report, 66% of law organisations are in the “developing” stage of AI maturity: teams testing proof of concept amid growing active use. Only 21% claim to be at a “mature” stage, actively using AI on client work and aggressively expanding its scope and use.  Firms are beginning to figure out that general-use LLMs aren’t enough to reach AI maturity, and products tailored to specific, well-worn internal processes are essential. For simple tasks like personal organisation and general fact-finding, generic LLMs function well. Under compliance pressure, having to navigate complex workstreams while keeping data completely private, they collapse. How could lawyers justify to clients their billable hours, the backbone of firms’ earnings that range between $500 and $1,500 per hour, if they use ineffective generic LLMs? The legal industry thrives on curated datasets, guardrails, and mind-numbing precision. Robust, “compliance-by-design” legal AI, moulded by strict governance, is the only way to operate. Regulatory hoops ensure companies never take a shortcut, even if the shortcut was just walking in a straight line. Battle-hardened tech So, what advantages does Europe have over its competitors in developing legal AI?  One: trust in the technology exists because it is built in a giant playground fenced by over 6,000 pages of legislative text. Beyond the AI Act, the EU’s General Product Safety Regulation (GPSR), which came into effect in December 2024, brought many AI-powered products within its remit, despite focusing on physical goods. Ensuring comprehensive user safety is paramount in the EU.  Noble as they might sound, high standards are maintained because EU law and regulation often scare off unserious startups (and some serious ones), or nefarious actors in the field. Clients valuing compliance will pay extra for tools that have the “We survived Brussels!” badge of honour. Two: the EU’s AI Act forces businesses to prioritise their competitive moats from day one, making them heavily armoured. The Act moves to establish regulations, identify high-risk AI systems, and create special provisions for general-purpose AI models. It distinguishes between AI systems that merely assist lawyers (limited risk) and those impacting the delivery of justice (higher risk).  Three: data rules, though a daily migraine for AI engineers, turn privacy into a selling point. GDPR’s “privacy by design” principle is intimidating for companies building outside the EU. But inside, businesses have already waded through the quagmire by the time the product reaches the market.  Europe’s regulation-first model could become the global template, or a cautionary tale. Only time will tell whether the Dodgeball logic of crossing the busy highway was the reason for victory or just an absurd rite of passage. In the end, the US and Asia might just let Europe do the exhausting norm-setting and then copy the good bits without the headaches. Yet while the rest of the world sees red tape as a nuisance, Europe’s legal sector sees it as a competitive track. In the global race, Europe’s advantage may not come from having the best tech. It could lie in having tech that can withstand the EU’s unique brand of “if you die in training, you live in competition.” source

The European space industry is booming. Yet despite the boom, the industry is struggling to find commercial buyers for arguably its most valuable output: data.  At the Living Planet Symposium 2025 in Vienna, the European Space Agency (ESA) and private sector leaders laid out Europe’s bold space ambitions and called for increased cooperation to address deep commercial gaps. Josef Aschbacher, ESA’s director general, highlighted one key focus. “Earth observation within the European Space Agency is a major priority,” he said. ESA has had recent successful missions. Its miniature satellite Φsat-2, for example, has started transmitting high-definition images back to Earth and will support wildfire, earthquake, and flood disaster management. The satellite can also be used to detect ships, gather data on illegal fishing, and monitor marine pollution.   However, for European startups looking to develop innovative new services from space, the industry can feel like a siloed bubble. Daniel Smith, Trade and Investment Envoy for Space for the Scottish Government and founder of AstroAgency, warned there are major fractures between the different players. He pointed to the disconnect between launchers, upstream and downstream space companies, and European businesses that’s preventing them from benefiting from space data.  “There’s still a lot of work to be done, because these companies are still struggling to commercialise,” Smith told TNW. “They’re still struggling to sell that data to other sectors. Because the space sector doesn’t want to buy the data.” Getting the message right This year, ESA began operations with a budget of €7.68bn (about $7.91bn). This represents a 1.4% decrease in funds from 2024, and is dwarfed by NASA’s $25.4bn allocation for 2025. It also fails to compete with the estimated budget of the China National Space Administration (CNSA).  To fill the budget gaps, maintain a competitive edge in space, and meet its ambitions, ESA has turned to the private sector. It aims to become a leader in Earth observation services, but can only get there by engaging with the local private sector, which, according to Smith, is out of the loop on uses for space technology.  This could be a big problem. If European companies do not create solid demand for space tech services, the entire industry will be put at risk. This includes spaceports, satellite manufacturers, and rocket makers.   “I’m seeing Earth observation companies going out of business, some of which are more than 10 years old, and they’re closing down because they can’t commercialise,” said Smith.  While government organisations like ESA offer grants and incentive programs, the applicants often don’t focus on monetisation and commercialisation, Smith added.  That leaves them missing out on big opportunities. Earth observation primarily utilises data from low-Earth obit (LEO) satellites, which have extensive use cases. LEO is already well-established in weather prediction and climate change applications. Sectors like agriculture, energy, infrastructure, logistics, maritime, and finance are also applying the satellite data to drive real-world impact. Adoption depends largely on leadership vision and data capability. Innovative use cases continue to emerge. For example, Scottish tech company Space Intelligence uses satellite data to create trust in the carbon offset financial market. The key to unlocking Earth observation’s potential, according to Smith, is reframing how we think about the tech:   “Space technology is ultimately about Earth, not about space.”  Selling space data The space industry is typically divided into upstream and downstream sectors. Upstream covers everything from manufacturing to launch, including rockets, spaceports, and satellite operations. Downstream services, meanwhile, offer ready-to-use space data for private companies. Downstream space data providers — driven by software developers and coding experts — retrieve satellite data, analyse it, and make it accessible for private companies. The data they collect in LEO orbits can be immensely valuable. It could drive the European space industry and advance the development of its spaceports, rocket launchers, and satellite companies. Currently, however, commercialising the data is challenging. As Smith explained, many space companies are struggling to sell their Earth observation data to other sectors.   “The space sector and rocket companies don’t want to buy the data,” said Smith. “They want to enable the data, they want to launch the satellites, so it’s still a big gap.” The reasons why European companies do not use space data in their operations are diverse. They include a lack of understanding of use cases, stigmas associated with space — such as slow and expensive processes — and the industry’s failure to open up and clearly communicate its value to potential private partners.  Where startups can start European companies considering Earth observation data for new ventures need to focus on downstream space services, Smith said.   Spire provides a positive example. The company operates a vast constellation network of affordable nano-satellites that operate in LEO, where they collect rich, granular data. The data has supported a variety of use cases. These include greenhouse gas emissions monitoring, IoT system optimisation in manufacturing, natural disaster monitoring, and maritime data analytics for commodity traders.  Spire breaks apart the upstream-downstream business model by taking a holistic approach to the entire space data supply chain. The company has manufactured and launched more than 200 satellites, building them in Europe and launching them from spaceports around the world.  “They focus very much on constellations to provide [continual] coverage,” Smith said. “They build the satellites for customers, but they build them for themselves as well, and then they sell their data that comes from their satellites,” he added.   Spire is not the only company providing innovative downstream services in Europe. Other examples include Catalyst, which recently signed ESA’s “Statement for a Responsible Space Sector,” and Hydrosat, which specialises in water, irrigation, and crop management solutions. Hydrosat’s latest satellite launched in June. It aims to advance the use of thermal satellite data and AI for food production, security, and natural resource management. Powerful use cases are also emerging in critical infrastructure and resource management. Forestry services, for example, are using space data from downstream providers to transform processes that were previously done manually. Tasks like inspecting and measuring forest health, size, and

The scientific world is reeling. New discoveries from the James Webb Space Telescope — a joint project by the European Space Agency(ESA), NASA, and the Canadian Space Agency (CSA) — aren’t just surprising, they’re contradicting our deepest assumptions about how the universe works. Fundamentally, it seems the universe may not be playing by the rules we mostly thought we understood.  So, what could it all mean for space exploration, space technology, and future deep tech? And what should space tech businesses, inventors, investors, and VC funds in Europe be considering as a result of the latest discoveries? At Beyond Earth Ventures, we’re all about startups building rockets, AIs for satellites, space biotech, and fusion breakthroughs.  But as space fanatics, we also like to look deeper, beyond cap tables and pitch decks, into the places where theory breaks and mystery begins.  TNW City Coworking space – Where your best work happens A workspace designed for growth, collaboration, and endless networking opportunities in the heart of tech. Enter the $10bn Webb telescope, sent into orbit from Europe’s spaceport in French Guiana, to look at the oldest light in the universe. Launched in 2021, the machine has been fully operational since July 2022. Webb isn’t just an upgrade from Hubble. It’s a time machine, an infrared sentinel, and — maybe most importantly — a destroyer of comfortable scientific assumptions.  Thanks to its findings, it’s becoming clear that we’re on the cusp of a major shift in theoretical physics and cosmology. Over the next few years, expect a wave of bold new theories, revisions to textbooks, and a renewed debate about everything from gravity to the origin of galaxies. Before we consider the implications, let’s zoom out and consider the big discoveries from Webb that punch holes in what we thought we knew about the universe. Some of these are already triggering theoretical crises. Others might trigger entirely new fields of inquiry and invention. The biggest revolutions start when theory no longer matches data. That’s what happened with quantum mechanics. With general relativity. With DNA. And maybe, with the Webb Telescope. Here are 10 of its discoveries challenging our theories about the universe: 1. The universe is expanding faster than it should We knew about the “Hubble Tension,” but Webb just confirmed it with more precision. According to the maths, the universe is expanding at 70–76 kilometres per second per megaparsec (km/s/Mpc) — much faster than the 67 km/s/Mpc predicted by models based on the early universe (the cosmic microwave background). Translation? Something in our physics is wrong, or at least incomplete. A tweak to dark energy? A new force? A misunderstood early universe? The door is open. 2. Galaxies grew up too fast Webb spotted fully-grown, massive galaxies just 500–700 million years after the Big Bang. These things are as large as the Milky Way, but their early appearance defies established science. According to the standard cosmological models, they simply shouldn’t exist yet. Theories say galaxies grow slowly. Reality says: they bulked up fast. Either we’re missing a trick — or the early universe was a lot more efficient than we thought. 3. Dark matter may be wrong — MOND was right? This one’s controversial: Webb’s findings align more with Modified Newtonian Dynamics (MOND) than the prevailing dark matter theory. MOND has long been the underdog of gravity theories. But if early galaxies are brighter and bigger than expected —  just as MOND predicted — we may need to reconsider which invisible hand is shaping the cosmos. 4. Black holes were way too ambitious How do you get a 9-million-solar-mass black hole only 570 million years after the Big Bang? That’s what Webb found. This is astonishing because, according to current models, there simply wasn’t enough time or material in the early universe to grow such colossal black holes so quickly — suggesting either unknown physics or entirely new formation pathways. The black holes in some early galaxies are 1,000x more massive (relative to the galaxy) than those in today’s universe. Either black holes formed via some exotic mechanism — or they started as something much bigger than stars.  5. Complex chemistry? This early? The galaxy JADES-GS-z14-0 is just 300 million years old, but it’s already rich in elements like nitrogen, which usually takes billions of years and several generations of stars to build up. How did those elements get there? Either the first stars formed and died much faster than we thought — or the Big Bang left us more “pre-built” than expected. 6. Stars formed at warp speed Webb shows early galaxies as intense, explosive star factories — a surprise to scientists. Models expected slow, gradual star formation. Instead, it’s “giant balls of star formation.” Something — perhaps a lack of dust, or different physics — accelerated the timeline. And, again, the models can’t keep up. 7. Planetary disks last longer than we thought Planet-forming disks around stars were assumed to vanish quickly. But Webb sees them lasting 20–30 million years. That’s great news for exoplanet formation — and potentially for life. If planetary systems have more time to develop, life-friendly environments may be more common than we ever dared to hope. 8. Galaxies were weirdly shaped Half of early galaxies look like pool noodles or surfboards, not the small round blobs we expected. The standard model says structure comes later. But Webb’s showing us that galaxies got organised early — and in shapes we weren’t expecting. Something about angular momentum and matter dynamics in the early universe needs rethinking. 9. Exoplanet atmosphere models are all wrong Webb’s ultra-precise spectroscopy revealed that our models of exoplanet atmospheres can’t reliably distinguish between different kinds. This shakes up everything from habitability estimates to the search for biosignatures. Basically, our “spectral fingerprints” are smudged — and it’s back to the drawing board. 10. The cosmic web was already there Webb found a 3-million-light-year-long filament — part of the cosmic web — just 830 million years after the Big Bang. This structure was supposed