Being AI-first isn't about giving employees access to AI tools. It's about rethinking how work gets done. I joined John Jantsch on the Duct Tape Marketing podcast to discuss how we're building an AI-native organisation at @Bookipi1, why the biggest shift is in mindset, not technology, and what it takes to build AI systems that actually deliver value. Have a listen: open.spotify.com/show/6DWKdS…

Oracle is targeting $225B in annual revenue by 2030. But there's another number worth paying attention to. The company doesn't expect to turn free cash flow positive until the second half of FY2028. AI infrastructure requires enormous investment before the revenue arrives. Growth and cash generation can follow very different timelines. As AI products move towards usage-based pricing, companies need to understand when infrastructure costs are incurred, when customers pay and how quickly that investment is recovered. A product can have excellent revenue growth and still create serious financing problems. The economics of AI-native software need to be designed around cash flow, not just ARR. investor.oracle.com/investor…

Google Search revenue just grew 12%, reaching $63.4B in a single quarter. Even more surprising, Google says search-query growth is accelerating at its fastest year-over-year rate in five years. So much for AI killing Google Search. But I think there's an important distinction for marketers. More searches don't necessarily mean more website visits. AI can encourage people to ask more questions while answering more of them directly. Google can grow its advertising business even if publishers receive a smaller share of that activity. In our team we are watching qualified traffic, customer acquisition costs and revenue from organic search rather than rankings alone. Google's business is growing. That doesn't guarantee the economics of SEO are improving for everyone else. sec.gov/Archives/edgar/data/…

The most expensive AI model shouldn't be the default for every business task. Liquid AI just released LFM3, a family of small, open-weight models designed for specialised work. They can be fine-tuned on a single GPU and deployed on company-controlled hardware. Consider the thousands of repetitive decisions inside a business: classifying support tickets, extracting invoice information or routing customer enquiries. Most don't require frontier-level reasoning. I'd benchmark smaller models against the best available models using accuracy, failure rates and cost per completed task. A cheaper model isn't a saving if someone has to correct half its work. For AI-native SaaS, model selection is increasingly a margin decision. liquid.ai/blog/lfm3

Creative production has an expensive problem: teams often discover a bad idea after they've paid to make it. Google's new Genie 4 lets people explore AI-generated environments in real time, changing camera angles, lighting, characters and dialogue as the scene unfolds. It's still an experimental model, but the implications for creative teams are significant. A director could explore several versions of a scene before booking a location, hiring a crew or committing to production. For marketers, the value isn't generating unlimited creative concepts. It's making better decisions before those concepts become expensive. deepmind.google/blog/genie-4…

China's AI bubble has burst. $480 billion wiped out in less than four months. The STAR50, China's index of technology and semiconductor leaders, is down 40%. Some of its most celebrated AI companies have lost 50% to nearly 90% of their value from their peaks. Meanwhile, the Nasdaq is down around 6% over the comparable period. There's a lot more to this story than a stock market correction, and I think it tells us something important about where the AI industry is heading. China spent years encouraging investment in AI, semiconductors and robotics. Government support gave investors confidence, capital flowed into technology companies, and valuations increasingly reflected expectations of future dominance rather than current financial performance. At one point, AI-related stocks represented approximately 42% of Chinese market capitalisation and 57% of trading turnover across mainland and Hong Kong shares. That level of concentration creates a problem. When everyone is investing in the same growth narrative, even companies executing well can become dramatically overvalued. But I think the financial correction could actually accelerate the next phase of AI adoption. 1. AI companies are discovering that technological leadership doesn't guarantee attractive economics. Training powerful models requires extraordinary capital. Operating them requires infrastructure, energy and continuous investment. And the competitive advantage of a new model can disappear surprisingly quickly. China is an interesting example because its model developers are competing aggressively on price. DeepSeek, Qwen, MiniMax, Z.ai and others are making increasingly capable models available at prices that would have seemed unlikely a few years ago. For customers, that's fantastic. For the companies financing the models, it's much more complicated. If model capabilities continue to converge while prices decline, the amount customers are willing to pay for access to intelligence could fall faster than the cost of developing it. That's a difficult business model, particularly when investors expect both rapid growth and improving margins. We saw a similar dynamic in other technology markets. Competition drove down the cost of infrastructure, while enormous businesses were eventually built by companies using that infrastructure. I suspect AI will follow parts of that pattern. 2. The AI industry is creating value faster than it can capture it. This is probably the biggest misunderstanding in the current debate. A technology can be incredibly valuable to the economy while producing disappointing returns for many of the companies financing its development. What happens when the cost of a capable AI model falls significantly? A business can automate more customer support conversations. Developers can run more coding agents. Marketing teams can analyse more data, generate more experiments and build internal tools that previously required dedicated engineering resources. The economic value created by those activities can increase while the price paid to the model provider decreases. As someone building and deploying AI agents inside a business, I find this particularly interesting. Our objective isn't to use the most expensive or technically impressive model. It's to find the most reliable way to complete a task at a cost that makes economic sense. If a cheaper model delivers the same business outcome, the savings can be reinvested into running more agents, automating more workflows or improving the customer experience. The benefit goes to the business using AI, even if the model provider captures less revenue per task. 3. I think we're going to see a redistribution of value across the AI industry. Today, enormous amounts of capital are being directed towards computing infrastructure, chips, data centres and foundational models. Some of that investment is clearly necessary. AI demand is growing and the infrastructure needs to support it, but the assumption that every layer of the industry will generate extraordinary financial returns deserves more scrutiny. As models become cheaper and more interchangeable, I expect more value to move towards companies that own distribution, customer relationships and the workflows where AI is actually used. A foundational model might be technically superior today and lose that advantage within months. A company with millions of customers, deep product integrations, proprietary business context and years of customer trust has advantages that are harder to replicate. That doesn't guarantee success. The application layer will also face intense competition. But it changes where I would look for sustainable businesses. The question becomes whether a company can turn increasingly affordable intelligence into something customers repeatedly find useful and are willing to pay for. 4. The real AI winners will have to prove their economics. Right now, the industry celebrates model benchmarks, funding rounds, infrastructure spending and the number of agents being deployed. Those metrics tell us something about progress, but they don't necessarily tell us whether a company has built a sustainable business. I'm much more interested in a different set of questions. - How much does an AI-powered task cost to complete? - How much measurable value does it create? - Does the customer continue using it after the initial excitement disappears? - Can the company improve its margins as usage increases? - Does the product become more valuable as customers integrate it into their daily work? For SaaS businesses, I think AI will force an important reassessment of pricing. Traditional subscriptions charge for access. AI introduces a meaningful variable cost every time a customer performs certain actions. That creates opportunities for usage-based and credit-based pricing, but it also introduces complexity. Companies need to understand whether increased usage improves profitability or simply increases their infrastructure bill. High usage without retention, pricing power or healthy margins isn't necessarily a successful product. The companies that understand these economics early will have an advantage over those optimising exclusively for adoption. 5. The US market shouldn't dismiss what is happening in China. The two markets are different. American AI investment is supported in part by extremely profitable technology companies with substantial existing cash flows. China's market has had more speculative participation in recently listed AI and semiconductor businesses. So I wouldn't expect the same correction to happen in exactly the same way. But there is a common question underneath both markets. Will the revenue and profits generated by AI justify the extraordinary capital being invested today? If AI becomes dramatically cheaper, adoption could increase enormously while the returns on some infrastructure investments fall short of expectations. That's a plausible outcome, and one investors should take seriously. It could also create an unusual situation where AI becomes one of the most transformative technologies in history while a large number of AI investments lose money. Those outcomes are perfectly compatible. 6. What I think comes next. I expect consolidation among foundational model developers, continued pressure on inference prices and more companies questioning whether they really need to build or own their own models. At the same time, cheaper intelligence should make previously uneconomical applications viable. Small businesses could eventually access capabilities that previously required entire departments. Software products could execute workflows rather than simply provide interfaces for employees to complete them manually. And businesses will increasingly select models based on cost, reliability, latency and the job being performed, rather than committing every task to a single provider. For companies building products, the opportunity is to turn this falling cost of intelligence into measurable customer value. For investors, the challenge is identifying which companies can actually capture that value. For AI labs, the challenge is maintaining a technological advantage long enough to build a sustainable economic one. I don't think China's crash tells us that AI has been overhyped as a technology. It tells us that investors may have overestimated how much of AI's economic value the companies building the underlying technology will be able to capture. And that distinction could define the next decade of the industry. The AI revolution can succeed spectacularly while many of today's AI investments fail. reuters.com/world/china/chin…

Anthropic just reported 900% net revenue retention. According to its IPO filing, qualifying existing customers spent roughly 9x more in the September quarter than they did a year earlier. About 80% of Anthropic's first-half revenue came from consumption-based pricing. That's a very different expansion model from selling additional seats. A company can keep the same headcount while multiplying its AI spending as more work gets automated. But there's a catch. As models become cheaper and more efficient, revenue per task can fall even as usage increases. For AI-native SaaS, I think the next important metric will be revenue expansion relative to the actual work completed. 900% NRR is extraordinary. The question is how much of that expansion survives falling prices per task. sec.gov/Archives/edgar/data/…

Google may eventually have to pay for content it previously received in exchange for traffic. For years, the relationship was straightforward. Publishers created content, Google indexed it, and search results sent visitors back to their websites. AI search is changing that arrangement. Google now has a content-licensing pilot involving 400 publishers and is reportedly testing payments for content used in AI-generated answers. I think this creates a fascinating imbalance. Large publishers have valuable content libraries and the ability to negotiate licensing agreements. Smaller businesses publishing content to acquire customers have considerably less leverage. For marketers, this makes the economics of SEO more complicated. Being referenced by an AI assistant might influence a purchasing decision without generating a measurable website visit. But visibility alone doesn't pay for content production. The next challenge for SEO teams is proving how content contributes to customer acquisition when the journey increasingly happens outside their website. theverge.com/column/1017983/…

I've helped grow businesses 10x, from startups to scale-ups and through to IPO. After 20 years in marketing, one thing I've learned is that building a successful business doesn't require a large team or an enormous marketing budget. Some of the biggest growth opportunities come from small teams that know exactly where to focus. And today, with AI, a solo founder can do things that would have required an entire team just a few years ago. Here are 11 growth principles I'd focus on if I were starting a business today. 1. Positioning beats promotion. Before spending a dollar on marketing, make sure people understand what your business does and why they should care. If someone can't explain your value in one sentence, more advertising won't solve the problem. 2. Talk to customers before building more features. A founder's biggest advantage is being close to customers. Talk to them. Understand what frustrates them, what they currently use and what they're willing to pay to fix. Ten meaningful conversations can be more valuable than weeks of building features nobody asked for. 3. Think about distribution from day one. A great product doesn't automatically attract customers. Before building, ask yourself how people will discover it. Can customers refer others? Can the product generate content that attracts new users? Can integrations or partnerships create distribution? The best acquisition channels often start with the product itself. 4. Your first 100 active customers matter more than 10,000 signups. Don't confuse attention with traction. Focus on getting a small group of customers to experience real value, return regularly and become willing to recommend the product. If people sign up but never come back, scaling acquisition will only make the problem more expensive. 5. Sell the outcome, not the technology. Customers don't care how sophisticated your AI agent is. They care whether it saves five hours a week, helps them win more business or gets their invoices paid faster. Make the outcome obvious. The technology is how you deliver it. 6. Retention is your cheapest growth channel. For a small business, replacing customers who leave is expensive. Focus on building something people continue to use and are happy to recommend. A handful of loyal customers who generate referrals can be more valuable than hundreds of users acquired through paid campaigns. 7. Price for value, not for being the cheapest. One of the biggest mistakes small businesses make is underpricing because they're afraid customers won't pay. A lower price isn't always a competitive advantage. Understand the problem you're solving, the value it creates and what customers are willing to pay. You don't need thousands of customers if you have a sustainable business serving hundreds exceptionally well. 8. Build repeatable systems before hiring more people. When something works, document it. Turn repetitive tasks into workflows. Automate where possible. Make sure the business doesn't depend on you doing everything manually. The goal isn't to create more work as you grow. It's to create a business that can serve more customers without increasing costs at the same rate. 9. Run small experiments constantly. You don't need a six-month marketing strategy to discover whether an idea works. Test a landing page, a different offer, a new audience or a new acquisition channel. Measure the results, learn and improve. For a small team, speed of learning is one of the few advantages you have over much larger competitors. 10. Use AI as a team multiplier, not just a writing assistant. A solo founder can now build systems for research, customer support, lead generation, marketing, reporting and business operations. But the real opportunity isn't generating more content. It's designing workflows where AI handles repetitive execution while you focus on customers, strategy and making better decisions. Think about which responsibilities you can delegate to AI before deciding which tasks require another hire. 11. Hire people who multiply your capabilities. Your first hires will shape the business far more than your twentieth. Look for people who take ownership, solve problems independently and bring skills you don't have. A brilliant individual who damages the team's culture can cost more than they contribute. The best small teams aren't made up of people who simply execute tasks. They're made up of people who make everyone around them better. If I were starting from zero today, I wouldn't try to build a large company immediately. I'd focus on finding a real problem, getting a handful of customers to pay for a solution, and building a repeatable way to acquire and retain them. Then I'd use AI and automation to remove as much repetitive work as possible. Only then would I start scaling the team, marketing spend and operations. The advantage of being small isn't just having lower costs. It's being able to learn, adapt and execute faster than companies ten times your size. And after 20 years, that's still one of the most powerful competitive advantages I know.

This month we have already beaten our efficiency goal by 17ppts. This is our playbook At @Bookipi1 , we build internal agents around real work our team needs to get done. 1. start with the work 2. define the outcome 3. build the harness 4. provide the right context 5. connect tools that can do the work 6. design the working loop 7. add memory and handoffs 8. human in the loop 9. test the real outcome 10. monitor and improve The goal isn’t to add more AI. It’s to give the team more time for work that needs human judgement.

The AI race is getting more entertaining than the products themselves.

ARR can keep growing while the underlying product is becoming less valuable to customers. Rahul Vohra shared something from Superhuman that I think more SaaS companies should adopt. Alongside ARR, they tracked "delighted ARR", revenue coming from customers who were genuinely engaged with the product. It's a useful distinction. An annual subscriber can stop using a product in February and still appear in ARR reports until December. By the time that revenue disappears, the product problem could be almost a year old. At @Bookipi1 we measure revenue alongside the frequency of completed customer workflows. Revenue shows what customers have agreed to pay. Consistent product usage offers an earlier indication of whether they'll keep paying. techcrunch.com/2026/10/23/vc…

An AI receptionist that sounds human but loses the booking is a very expensive demo. @jack's Block just unveiled Dessa, a voice model trained entirely on human conversations, without using text. It can listen and speak simultaneously, handle interruptions and run locally on a MacBook. It's still a research preview, but I think the direction is important. Real customer conversations are messy. People interrupt, change their minds halfway through an order, ask unexpected questions and expect the conversation to continue naturally. For a restaurant, the value isn't how realistic the voice sounds. It's how many reservations are completed, how accurately orders are taken and how much each successful interaction costs. The breakthrough for voice AI will be making those conversations reliable and economically viable at scale. block.xyz/dessa

Google's next advertising advantage could come from the personal information people share with AI assistants. Its updated privacy policy allows information shared from Gmail, Drive and Photos with connected Google services to influence ad personalisation, depending on user settings. The connections are optional, and advertisers don't get access to personal inboxes or files. But I think this raises a much bigger question about the economics of AI assistants. The more an assistant understands someone's preferences, plans and purchasing intentions, the more valuable that context becomes for advertising. Google has an enormous advantage here because it already owns so many of the services people use every day. The challenge will be maintaining trust when information shared to receive a better service can also influence commercial recommendations. Consent is becoming part of the competitive advantage in AI. policies.google.com/privacy

Canva just made a move that most SaaS companies can't afford to make. It built its own AI image model. Craft is trained around actual design templates, typography and layouts, with outputs that can be edited inside Canva. But I think the biggest advantage could be financial. With more than 250M users, the cost of generating images becomes a significant consideration. Every improvement in inference efficiency can have an enormous impact at that scale. Owning the model also means Canva can optimise specifically for what its customers need, rather than relying on another company's roadmap. There's a broader lesson for AI-native SaaS. Adding AI capabilities is becoming increasingly easy. Building a product where the economics improve as usage scales is considerably harder. canva.com/newsroom/news/canv…

Walmart is opening its doors to AI shopping agents while Amazon is trying to keep some of them out. Walmart just introduced dedicated agent profiles, spending limits and an API-based gateway for AI assistants to shop on behalf of customers. I think the difference in strategy is fascinating. Amazon has spent decades building a highly profitable ecosystem around product discovery, advertising and purchases. Allowing external agents to control more of that journey could threaten parts of its economics. Walmart appears more willing to let the customer relationship start elsewhere, provided the transaction still happens on its platform. Both approaches make commercial sense. The question is whether AI shopping creates genuinely new demand or simply moves existing purchases into an interface controlled by someone else. Retailers might gain sales while losing influence over which products customers discover. corporate.walmart.com/news/2…

The next big revenue opportunity for vertical SaaS might have very little to do with software subscriptions. Slingshot just raised $27M to make embedded insurance commercially viable, including policies worth as little as $25. Offering relevant insurance during checkout creates another revenue opportunity without needing to acquire another customer. The difficulty is that servicing a $25 policy can involve almost as much administrative work as a much larger one. Slingshot is using AI and automation to reduce that overhead. I think this is an underrated opportunity for vertical SaaS. The most valuable expansion products may be financial services, payments and other transactions that naturally fit into workflows customers already use. The challenge is finding enough margin after operational costs, rather than simply adding another revenue stream. news.crunchbase.com/fintech-…

Microsoft is shutting down The Garage after 17 years. The program gave employees space to experiment, build prototypes and collaborate on projects outside their usual work. Microsoft says experimentation now happens inside everyday teams. I like the ambition, but there's a risk in assuming innovation will naturally happen once it becomes everyone's responsibility. A normal product roadmap is designed to deliver agreed priorities. Experimentation requires room for ideas that might fail, don't have an immediate ROI and sometimes have nothing to do with the current roadmap. AI makes building prototypes significantly easier. But it doesn't automatically create the organisational freedom to pursue them. If innovation is going to happen inside every team, employees need the authority to experiment, not just access to better tools. Otherwise, innovation becomes everyone's responsibility and nobody's priority. geekwire.com/2026/microsoft-…

Grok's new Pages feature could turn X into something far more valuable than a social network. The rollout lets Grok generate editable, automatically updated reports and dashboards, using X's real-time information. I think the strategic opportunity is bigger than dashboards. X already has a constant stream of information from founders, developers, customers and industry experts. Most of that information disappears into a feed within hours. Turning it into persistent, personalised tools changes how people can use the platform. A marketing team could follow changing customer conversations and unlike a traditional dashboard, the output can evolve as new information appears. If X can connect real-time conversations with analysis and execution, it could become a serious business intelligence platform. The real competitive advantage is owning the information that keeps it useful.

Google is about to start selling OpenAI models to its own cloud customers. I think this is a smart move from both companies. For OpenAI, it removes a significant barrier to enterprise adoption. Customers can access its models through infrastructure, security controls and procurement processes they've already approved. For Google, it reinforces the value of its cloud platform, regardless of which AI model a customer prefers. There's a lesson here for SaaS companies. Having the best product doesn't guarantee distribution, particularly when customers already have established vendors and budgets. Sometimes the fastest way to grow is to make your product available where customers already spend their money. openai.com/index/openai-mode…