Disruptive Intelligence
Every week, we filter the AI chaos to bring you fresh innovation insights and actionable takeaways.
Top Insights
1. Treating AI as a normal technology is counterproductive
Firms are treating AI like standard enterprise software, embedding it into workflows, assigning KPIs, and handing it to IT. This instinct, normally good management practice, is exactly wrong for AI. This turns a transformative capability into incremental efficiency. For instance: companies can choose to use AI to summarize meetings, instead of experimenting using AI to redesign how decisions get made (e.g., AI-generated scenarios replacing meetings entirely).
Leaders see productivity gains (e.g., +30%) and jump to cost-cutting (e.g., -30% headcount). But the easier path (cutting costs) is strategically inferior to the harder one (reimagining output). AI’s real leverage is not doing the same work cheaper, it’s enabling 10–100x output per expert, entirely new offerings, and expansion into adjacent markets.
AI requires experimentation, as well as the tolerance of failure and ambiguous outcomes. IT is often optimized for stability, control, and risk reduction. Putting AI in IT leads to safe, low-impact use cases but possibly kills innovation early.
🎯 What to do:
Have the CEO and business-unit leaders own AI strategy, with a 3-year mandate to define how AI changes the company’s products, economics, and org design, not just its efficiency.
Create a central AI lab with both technical and non-technical talent, and give each function executive sponsors and priority use cases tied to growth, speed, decision quality, or new offerings.
Make experimentation safe, visible, and rewarded: light guardrails, simple sharing mechanisms, and explicit assurance that useful AI adoption will not be punished or automatically converted into headcount cuts.
Reference: The IT department: Where AI goes to die (The Economist)
2. AI is disrupting business models
AI is becoming the first point of decision-making, not your website or sales funnel. Customers now ask AI tools (ChatGPT, Perplexity) for recommendations instead of browsing options themselves. The AI filters, compares, and decides what gets seen. For instance: instead of a customer Googling “best CRM” and seeing 10 vendors, an AI might recommend just 2–3 options. If you’re not included, you effectively don’t exist.
Entire layers of value chains are at risk of disappearing. AI removes intermediaries that aggregate information, compare options, and produce standardized outputs. Iff your business summarizes data, generates reports, or compares vendors: AI can now do that instantly and cheaper.
Barriers to entry are collapsing. AI dramatically reduces the cost and complexity of building a company. Small teams can now do what required large organizations. Expertise, coding, and marketing are increasingly automated. A 5-person AI-native startup can now compete with a 100-person company.
Customers are no longer willing to pay for effort. AI makes many outputs (content, code, analysis) cheap and fast. Time-based pricing (hourly billing, per-seat SaaS) becomes harder to justify. Customers expect outcomes, not effort.
🎯 What to do:
Audit how your company appears in AI tools (ChatGPT, Perplexity, etc.). Structure your content for machine interpretation (clear comparisons, structured data, APIs). Invest in third-party validation (reviews, benchmarks, datasets AI trusts).
Being an intermediary without a defensible asset is dangerous. Instead of selling reports, integrate them into decision systems. Instead of comparing vendors, become the infrastructure vendors plug into.
Identify the real outcome customers care about. Tie pricing to performance (e.g., revenue generated, cost saved), continuous value (subscriptions tied to results), and usage tied to impact, not seats.
Create a “shadow org”: small, AI-first team who rebuilds one product or function from scratch. Benchmark cost, speed, and output vs your core business.
Reference: Three structural ways AI is breaking your business model (Board of Innovation)
Innovation Radar
Researchers published a new AI evaluation method using general scales to quantify AI capabilities across tasks. Instead of isolated benchmarks, it profiles task demands and AI strengths to predict performance on new challenges.
→ This promises more reliable AI assessment: it will give organizations clearer insight into which AI systems truly have the needed skills (e.g. reasoning, coding, planning) for business use-cases, reducing the risk of investing in models that underperform in the real world.
OpenAI updated its ChatGPT-integrated apps (Box, Notion, Linear, Dropbox) to add new AI actions including content generation (“write” capabilities). Now ChatGPT can draft or modify documents, notes and project entries within those apps.
→Tightening integration with enterprise tools lets companies automate routine content tasks (e.g. drafting docs or updating tickets), boosting productivity and saving manual effort.
Google DeepMind released Gemma 4, a family of four open-weight LLMs (2B–31B parameters) under an Apache 2.0 license. These include ultra-efficient edge models (E2B, E4B) and large 26B/31B models for agentic reasoning, all supporting multimodal inputs (text, image, audio) and function-calling.
→ Firms should note that Gemma 4’s permissive license means companies can integrate powerful AI into their products at no cost, spurring innovation in areas from customer service bots to on-device analytics.
Google launched Veo 3.1 Lite, a lower-cost video-generation model accessible via the Gemini API. Veo 3.1 Lite offers the same low-latency generation as the flagship model but at roughly 50% of the cost, producing HD (720p/1080p) video clips up to 8 seconds with support for cinematic commands. Runway released Gen-4, a new generative model for images and video. Gen-4 can “precisely generate consistent characters, locations and objects across scenes” given a single reference, maintaining coherent style and physics across frames. It enables controllable video creation without fine-tuning. ByteDance rolled out Dreamina Seedance 2.0 in CapCut, an AI model that generates short videos (with audio) from text or image prompts. It can create up to 15-second clips (e.g., vertical 9:16 format) complete with AI-composed music, all without needing reference footage.
→ Video creation is becoming programmable, cheap, and of high quality, meaning competitive advantage will shift from production budgets to speed, narrative control, and distribution strategy.
Grateful you’re here. If this sparked something for you, pass it along: good ideas grow when shared.