So yeh. I attended Web Summit in Lisbon again in November 2024, returning to Europe’s largest tech conference.
Core Thesis
The conference surfaced a consistent pattern: value is shifting from production to orchestration. Whether it’s managing AI agents, allocating attention across workflows, or positioning platforms as trusted intermediaries—the winners will be those who coordinate, not just create.
A deeper pattern underneath: the shift from stock to flow.
- Websites → feeds → real-time intelligence
- Specialists → on-demand expertise
- Configuration → conversation
- Products → services → outcomes
- Content → context → intent
Traditional platforms have been "stock" products—you build something, it sits there. The future described here is all flow: continuous adaptation, real-time response, dynamic orchestration.
These patterns are both ultimately about dynamic resource allocation—time, energy, capital, attention, skills. As the pace of change accelerates, optimization becomes the differentiator. Those with clear domain models, effective coordination, and decision-making clarity will outperform those without.
The Mental Models Behind This Analysis
I’m synthesizing several frameworks to understand these patterns:
-
First Principles Thinking – Strip away current execution paths to understand fundamental user intent. What are people actually trying to accomplish?
-
Clayton Christensen’s Disruption Theory – But instead of asking "what gets disrupted," asking "in what sequence and why does order matter?" (Industrial → Service → Home rollout pattern)
-
Systems Thinking – Seeing the flow of value rather than static products. Stock vs Flow paradigm shift.
-
Value Migration (Adrian Slywotzky) – Tracking how value moves from production to orchestration as interfaces standardize
The synthesis reveals something none address alone: as change accelerates, competitive advantage shifts from what you can produce to how well you can orchestrate.
Day 1: Foundation Sessions
Building Europe’s Global Tech Edge
Speakers:
- Ling Ge, Chief Investment & Strategy Officer, EMEA, Tencent
- Archie Hollingsworth, Co-founder, Fyxer
- Arjun Kharpal, Senior Technology Correspondent, CNBC
Key Insights:
Ling Ge’s Investment Framework: Look at investments through three lenses:
- People – Team capabilities and dynamics
- Problem – Is this an enduring (10+ year) problem worth solving?
- Perspective – Unique angle or insight
Retention as the Critical Metric: App retention remains the hardest problem. User acquisition is solvable; keeping users is not.
Model Development as Moat: Leaders are bringing model development in-house. That’s where the moat is—not just using third-party models, but customizing for specific domains.
Not Always About Big Models: Relevance matters more than size. Tone, context, and domain specificity create competitive advantage.
"Vibe Revenue": Currently seeing lots of big raises without much to show. Funding doesn’t equal product-market fit.
From Plundering Corporations to Pillaging the State
Speakers:
- William Lazonick, Professor of Economics Emeritus, University of Massachusetts
- Kenneth Cukier, Deputy Executive Editor, The Economist
Focus: Convergence of corporate power and political control. How corporate strategies of value extraction—perfected through decades of shareholder primacy—are now being deployed against democratic institutions themselves.
Working It Out: Job Security in the AI Era
Speakers:
- Pavlina Tcherneva, President, Levy Economics Institute of Bard College
- Cory Doctorow, Author, Pluralistic (runs on WordPress!)
- Katie Collins, Principal Writer, CNET
Doctorow on Platform Decay ("Enshittification"):
Platforms used to have:
- Competitors – until antitrust was destroyed
- Regulators – until they were disbanded
- Open protocols – until IP moats and locked-down networks took over
- Empowered workforces – until supply caught up with demand
C-Suites typically put too much faith in AI capabilities.
Main Threat to Jobs: Financial speculation, not just automation:
- Electricity costs rising
- Health costs rising (more treatments possible but more expensive)
- Cost-adjusted wages falling
Why: When financial speculation becomes the main driver of profits, the real economy gets squeezed. Capital flows into assets instead of productive investment, firms extract rents instead of innovating, essential services become financial vehicles, and wages stagnate because labor becomes a cost to minimize rather than a resource to grow.
Algorithmic Discrimination: Contractors and gig workers create "per-worker wages" enabling cost competition per role. AI increasingly enables algorithmic discrimination:
- No new consumer privacy laws since 1998 in the US
- Massive data on potential employees/contractors available
- Example: Nurses with more debt offered lower wages ("desperation tax")
My Reflection: This talk was somewhat depressing and felt more like problem identification than solution exploration. Doctorow’s book "The Internet Con: How to Seize the Means of Computation" explores these themes, naming the problem well but solutions remain elusive.
The Allocation Economy: What Comes After Knowledge Work
Speaker: Dan Shipper, Co-founder & CEO, Every
Premise: When machines can write, code, and design at scale, human value shifts from producing to allocating—deciding what to build, how to shape it, and where to direct it.
Key Insights:
Everyone Needs to Be a Manager: To manage agents, workflows, processes. But machine management won’t be like human management.
Superpowers for Generalists: Historically, work rewarded specialists due to skill & experience scarcity. AI creates "good enough" specialists on demand, enabling generalists skilled at employing AI specialists.
AI Enables Creative Work with Fractured Attention: Definitely aligns with my lived experience—I spend less time in deep focus and more in context-switching and task management.
What Should Kids Learn?: Learn how to use AI and follow your curiosity.
Rate of Change: Will be slower at big companies, so we’re likely not seeing most of the impact yet.
CEO Usage Critical: Key is for CEO to use AI heavily to understand nuance and use cases.
Workforce Adoption Expectations:
- 10% early adopters
- 80% adopt when told/shown
- 10% highly skeptical
- Success = enabling the 10% to lead the 80%
More Managers, Flatter Hierarchy: Managing tools not people. AI is excellent at synthesizing and communicating information (e.g., AI meeting note tools like Granola).
My Reflection: I’m coincidentally reading "Range: Why Generalists Triumph in a Specialized World" by David Epstein, which discusses generalist advantages from a pre-AI era. Viewing it with the benefit of living in the AI age adds interesting context.
Machine intelligence will solve problems/behave in ways utterly unfamiliar to humans. This speaks to the importance of understanding first principles of an endeavor rather than specific execution paths.
From Demos to Deployment: The Enterprise AI Journey
Speakers:
- Jesse Zhang, Co-founder & CEO, Decagon AI
- Timothy Young, CEO, Jasper
- Anu Bharadwaj, President, Atlassian
- Mike Butcher, Founder and Editor, Pathfounders
Jasper: In 25% of Fortune 500 with enterprise licenses (impressive!)
Content Workflows: Updating thousands of product images and descriptions in hours, not months.
Atlassian Rovo: 1.4M monthly active users running 33k agents. All Atlassian users have access to Rovo because they now only offer managed platform (no more self-hosted).
Importance of Precision and Accuracy: Careful control of what data models have access to.
Organizational Rethinking: E.g., Marketing needs engineers to connect data, models, and workflows.
Chatbot Metric: Get "ask for human" rate as low as possible.
Cultural Change Themes:
- Atlassian (15k people): Internal efficiency – start simple (end-to-end workflow, clear benefit)
- Mercedes (35k people): 85% reduction in duplicate tickets
Atlassian Staffing: Let go 150 support staff – "internal readjustment," not directly AI replacement but AI enabled the change.
Configuration Becoming Intuitive: E.g., Salesforce has a whole industry of experts—AI will bring end customers into more direct contact with their tools (Bret Victor’s "Inventing on Principle" vision).
Opportunity: Be more human, offload technical work to AI.
My Reflection: This discussion covered the tools being built, how we work together and manage change, platform vs self-hosted implications, and distinguishing work that machines should do vs. work that humans should do.
Redefining Robotics with Boston Dynamics
Speaker: Robert Playter, CEO, Boston Dynamics
Spot (the robot dog) was on stage—very cool!
Training Data from Humans: Using humans with sensors to generate training data created great results.
Rollout Path: Industrial → service → home
- Controlled environment first
- Affordability comes from efficiency and scale (large industrial purchases)
4 Pillars:
- AI
- Customer value (application layer)
- Dependability
- Safety
Spot with Specialist Sensors: Inspects equipment on "smart rounds" autonomously.
Philosophy:
- "This is not a job that people should be doing"
- "Pause when operating near humans"
- The importance of non-verbal communication
My Reflection: SUCH a great talk, and not just for the robots on stage. The "smart rounds" concept is fascinating—what sensors would a digital inspection agent need? How can AI "pause around humans" in digital environments? What non-verbal communication cues would it look for?
The industrial → service → home progression is instructive: go where variance is lowest and stakes are clearest.
Day 2: Intelligence and Infrastructure
The AI Browser Wars
Speakers:
- Laura Chambers, CEO, Mozilla
- Matthew Prince, Co-founder & CEO, Cloudflare
- Mike Butcher, Founder and Editor, Pathfounders
Matthew Prince on Content Scraping: Perplexity behaving poorly regarding content scraping.
Trusted Intermediaries: Mozilla positioning as trusted intermediary (Apple too).
Importance of the Memory Layer: Whoever owns persistent context across interactions—user preferences, history, relationships—becomes the intermediary that AI systems have to route through.
OpenRouter’s play is model routing. The equivalent for content would be context routing: who holds the graph of what this person/business has created, who they’ve interacted with, what they care about?
Privacy-Preserving Personalization: Opportunity to abstract personal data into mathematical abstracts to pass to LLMs without exposing raw personal information.
Changing User Behavior: Fewer people are clicking on source links in AI-generated responses.
Cloudflare’s Position: Sits in the middle of publishing, users, and AI companies.
Critical Question: "Is this a human?" is increasingly important to answer.
Business Model Shift: If there are AI browser wars, who is writing the rules? The business model of the internet is about to change drastically.
Content Economics: Google gets content for free vs. OpenAI pays for it. Longer term, we either start charging Google to crawl sites, stop charging OpenAI, or find a more universal solution.
My Reflection: Cloudflare and Mozilla seem like natural allies for the open web. Their positioning as trusted intermediaries in the AI-mediated future is strategic.
The Intersection of AI and Robotics
Speaker: Tye Brady, Chief Technologist, Amazon Robotics
Physical AI: Machines "feeling" physically, not just computing digitally.
DeepFleet: Amazon has >1M robots in service powered by DeepFleet—trained by years of robot data (stops, movements, patterns). Can auto-reroute around bottlenecks → 10% efficiency savings.
Amazon’s edge: experience and data of machines in the real world.
5 Trends:
-
Physical AI learns from the grammar of movement – Motion, friction, flow. "Tokens to torque."
-
Mobile manipulation – "Chords not notes." Moving objects as they themselves move (Vulcan robotics & Bluejay systems). Touch & dexterity. Bluejay has multiple arms in a single system. Amazon operates in real-world contexts, not staged environments.
-
Build them cheaper, faster, better – Once they work in the real world, you need to make them affordable. Robotics for all. 3D printing, digital twins, AI all accelerate iteration from months to days.
-
Conversational robots – Robots that understand you, not just blindly completing commands. Project Luna: Intent not instructions.
-
Work in the future – People + machines. Extend and amplify what matters to you. Augmentation not replacement. Expand capability and increase access.
My Reflection: Reinforces the humans + machines, augmentation, management themes running throughout the conference.
Strategic Patterns Across Sessions
Several recurring themes emerged that form a coherent picture of technological transformation:
1. Technology Continues to Disrupt—Ever Faster
Financialization of the economy (activity → data → corporate capture → AI) is pulling value from the real economy into financial markets. This decorrelation between humans and value drives increasing shareholder power and corporate influence on government. Platform decay accelerates as checks and balances erode.
AI’s effects are still in their infancy: we are truly in what Amy Webb calls "The Beyond" – the phase where technology shifts from novelty to fundamental infrastructure.
2. First Principles Matter More Than Execution Paths
Understanding what users are actually trying to accomplish becomes critical when AI can execute any path.
Historical frame: People connected with people and companies through content distributed on the web through websites.
Current frame: People connect through individual content posts distributed on platforms through algorithmic feeds within closed apps.
Future frame: People connect through ideas (multimodal content) distributed through an intermediate network of intelligence in real-time global intelligence.
3. Models as Moats—Through Intent, Not Imitation
The competitive advantage isn’t model size; it’s fit. Tone, context, domain specificity.
But fit doesn’t mean training on what users currently build (often suboptimal). It means understanding what users want to achieve and helping them build it in a way that’s actually good—meeting functional and non-functional requirements they may not even understand or articulate.
This links directly to intent-not-instructions: infer the goal, execute it well. The model advantage is in the translation layer between messy human intent and quality output.
4. The Memory Layer as Strategic Ground
Whoever owns persistent context across interactions—user preferences, history, relationships—becomes the intermediary that AI systems have to route through.
OpenRouter’s play is model routing. The equivalent for content would be context routing: who holds the graph of what this person/business has created, who they’ve interacted with, what they care about?
5. Controlled Environments First
Boston Dynamics’ rollout path (industrial → service → home) is instructive. They went where variance was lowest and stakes were clearest.
Identify high-value, controlled, repeatable environments before expanding to high-variance scenarios.
6. Self-Hosted AI Challenges
Atlassian EOL’d on-prem solutions (ending 2029) partly because they couldn’t make AI perform well with that much environmental variance.
If that logic holds, it has serious implications for self-hosted software’s future—or at least for where AI features can credibly live.
This creates tension between open-source freedom and managed platform value creation.
7. Work Machines Should Do vs. Work Humans Should Do
Whether it’s Salesforce configuration or warehouse robots, there’s an opportunity to offload technical work and be more human.
Machines should: Use complex interfaces, select and configure tools, handle integrity testing, synthesize information across systems.
Humans should: Build relationships, innovate, make judgment calls, handle ambiguity, be humans together.
This isn’t about replacement—it’s about augmentation and appropriate allocation. The "pause around humans" principle from Boston Dynamics applies: AI should know when to step back and let humans just be.
8. More Managers But Flatter Hierarchy
Everyone needs to manage agents, workflows, processes. But machine management isn’t like human management.
AI is excellent at synthesizing and communicating information around organizations (previously the domain of middle management).
For those who don’t suffer from context switching, the bottleneck rapidly becomes the ability to run concurrent work efficiently across multiple branches/workstreams.
9. The 10-80-10 Adoption Reality
Dan Shipper’s framework:
- 10% early adopters
- 80% adopt when told/shown
- 10% skeptics
Success = enabling that first 10% to pull the middle along.
Distributed/async culture works against this because empathy and mindset enablement happens in proximity—the day-to-day version of pair programming.
Cultural change requires building trust and starting simple: end-to-end workflows, clear benefits.
Question: How do you create density for your 10%? Dedicated cohorts? Paired workflows?
10. Intent Not Instructions
Amazon’s Project Luna—robots that understand intent, not just commands—represents a fundamental shift in interface design.
Interfaces (human or agent) that infer what you’re trying to accomplish rather than executing discrete commands.
This closes the loop with model fit (#3): understand intent, translate to quality execution, learn from context (#4).
11. Physical Intuition as a Design Pattern
Boston Dynamics’ "pause around humans" and Amazon’s "tokens to torque" (physical AI learning the grammar of movement) have software analogues:
- When should AI tools "pause" during critical operations?
- What’s the "grammar" of a healthy digital system vs. a struggling one?
- Can you learn movement patterns, not just static metrics?
It’s not just monitoring—it’s developing "physical intuition" for digital environments.
12. Platform Decay and the Trusted Intermediary Question
Doctorow’s framework: platforms used to have competitors (until antitrust was destroyed), regulators (until disbanded), open protocols (until IP moats took over), empowered workforces (until supply caught up with demand).
Remove those checks and incumbents act purely in shareholder interests.
Mozilla and Cloudflare are positioning as trusted intermediaries. What does "trusted intermediary" mean in practice?
13. Holistic Business Solutions, Not Components
The vast majority of the expo hall was selling integrated, holistic business solutions. The market wants houses, not lego bricks.
This connects to ongoing disruption: as incumbents consolidate and platforms decay, the "just plug things together" model gets harder to sustain. Users want outcomes, not assembly projects—at least the mass market is telling us that’s what they want.
14. Generalists + AI Specialists = New Leverage
If AI creates "good enough" specialists on demand, the scarce resource becomes the generalist who knows which specialists to invoke and how to integrate their outputs.
Implications:
- Hire for breadth, train for orchestration
- Career paths shift from deepening expertise to expanding range
- "T-shaped" becomes "comb-shaped"
Expo Hall: Industry Landscape
The expo hall showcased the current state of the startup and enterprise ecosystem. Some notable categories and companies:
E-commerce & Payments
- VTEX – The Backbone for Connected Commerce – Enterprise e-commerce platform for omnichannel
- Payabl – Flexible payments infrastructure
- Productflow – E-commerce & retail platform
- Clarity Global – Cross-border e-commerce and payments
- SumUp – Payments including POS
- Visa – Major payment processor
AI & Development Tools
- ZenCoder AI – AI coding agent with built-in intelligence, devops, integration
- Locofy.ai – Design to code conversion (Figma → production code)
- Replit – Apps, Agents, Automation – Cloud IDE
- Contentful – Headless CMS
Cloud & Infrastructure
- Alibaba Cloud – Cloud computing with Tongyi AI Models (300+ models, 600M users, 119 countries)
- Pantheon.io – Cloud platform for web frameworks
- Kinsta – Premium WordPress hosting
- SaveOnSaaS – SaaS cost optimization
- Vercel – Frontend cloud platform (Next.js creators)
- Atlassian – Jira, Confluence, Trello
Marketing & Growth
- Awin – Affiliate marketing platform
- SocialBee – User engagement and growth
- SendPulse – Marketing & sales automation
- Infobip – Customer engagement with agentic AI
- Scalable Agency – AI agents for marketing
Developer Tools & Content
- Reply200 – AI-driven conversation automation, audience integrity
- FeedMug – Content management
- Social.plus – In-app community building
Financial Services & Compliance
- UMCA Tech – Financial fraud prevention
- Stape.io – Global payroll and contractor payments
- Globalfy – International company formation
Quality Assurance & Accessibility
- TestDevLab – Software QA
- Deque – Digital accessibility standard (675,000+ installed sites)
Niche Solutions
- AnimalHotels.com – Travel & hospitality for pets
- Yatta.de – End-to-end commerce for digital products
- Legacy.app – Digital legacy management
Overall Impression: The expo felt busy and engaging. Better-designed booths with cool demos, prizes, and exciting products saw significantly more traffic. The audience is building products and businesses—founders, developers, partners, ecosystem players.
Meetings & Conversations
Damarys Zampini (met at Algarve Tech Summit the week before):
- Just started using Odoo CRM – invoicing, doc signing, calendly
- Speaks to the ‘holistic business solution’ theme
Alex Dorweiler (Atlassian):
- Datacenter EOL 2029—no more self-hosted solutions supported
- Gone all-in on managed platform
- One strong reason: couldn’t make on-prem AI work well—too much variability
- This is for their Rovo product
Andy Budd (ClearLeft Founder, Seedcamp VC):
- Discussed "Finance as Air Traffic Control" concept
- Coordination and governance models for complex organizations
Key Questions for the Future
-
First Principles: In an AI-mediated world, do traditional content platforms still provide strong identities, clear relationship value, powerful distribution, and trusted intermediation? Is that fixable, or does it require a positioning pivot?
-
Stock to Flow: What would traditional content platforms look like if designed for flow, not stock?
-
Self-Hosted AI: If early-stage AI doesn’t work as well in high-variance self-hosted environments (Atlassian’s logic), what does that mean for where AI features can credibly live in open-source ecosystems?
-
Open Source + AI Tension: Can we synthesize (rather than compromise) between open-source freedom and managed platform value creation? Is there a solution that creates a clear strategic moat?
Reflections
Web Summit 2025 captured a fascinating moment: AI is maturing from demos to deployment, robots are moving from labs to factories, and the fundamental economics of the internet are shifting.
The Shift from Stock to Flow is perhaps the most profound pattern. Traditional software created stock—artifacts that sit there. The future is flow—continuous adaptation, real-time response, dynamic orchestration.
Value in Orchestration over production means the winners won’t be those who create the most, but those who coordinate the best. Managing agents, allocating attention, routing context, positioning as trusted intermediaries—these become the differentiators.
Dynamic Resource Allocation as the underlying mechanism ties it all together. As change accelerates, optimization becomes everything. Clear domain models, effective coordination, and decision-making clarity become competitive advantages.
The conference left me with more questions than answers—but the right questions. How do we design for flow? What does orchestration look like at scale? Who becomes the trusted intermediary? Where does value accrue in an AI-mediated future?
These aren’t just technology questions—they’re economic, social, and political questions about the future we’re building.
Event: Web Summit 2024
Location: Lisbon, Portugal
Dates: November 11-14, 2024
Published: November 27, 2024
Related Content:
- SXSW 2024: Technology Supercycles and AI Transformation – Earlier AI transformation insights
- Intent-Driven Commerce – Technical implementation of intent-driven systems
- There Is No Host – Platform abstraction and freedom stacks
Source: Originally published November 2024 as internal conference notes, sanitized and enriched for public release December 2024.