Think Big Newsletter #18 - Have Backbone, Designing AI UX, Shape of AI, and MCP Apps
AI, Innovation and Business Value
Welcome to the Think Big Newsletter, where we explore how business leaders can create value with AI and innovation. In each issue, I’ll share practical strategies, real-world case studies, and actionable frameworks that help you navigate the AI wave without getting lost in the hype.
I’d love to hear your thoughts, challenges, and suggestions for future editions.
In this issue:
Leadership Principles in the age of AI - Have Backbone…
Business Value with AI - Designing AI UX
AI Tool Deep Dive - Shape of AI
One New Term at a Time - MCP Apps
#1 - Leadership Principles in the Age of AI
Have Backbone; Disagree and Commit
I was at Tech Arena last week. This is one of the biggest tech events in the Nordics - a place to catch the leading trends, talk to startups, investor, politicians, and users. Almsot needless to say, AI was top of the agenda. Some difficult questions were being aksed, but maybe not enough. Maybe there was too much of a positive vibe.
It made me think on one of the most interesting and multifaceted Leadership Principles. Have Backbone, Disagree and Commit. It reads like this.
Have Backbone; Disagree and Commit.
“Leaders are obligated to respectfully challenge decisions when they disagree, even when doing so is uncomfortable or exhausting. Leaders have conviction and are tenacious. They do not compromise for the sake of social cohesion. Once a decision is determined, they commit wholly.”
It is an especially important principle when everyone seems to think alike. When the culture is that of acceptance and even consensus (like is also a common belief about Sweden - where I live and work).
We can think of this principles in many different ways in the age of AI. I’ll touch on some of them, and will revisit it in future issues.
The age of AI obligates us to rethink common beliefs
The abundance of intelligence is undermining many of the things we have been taking for granted. That things have to be done in certain ways. That some things are more important than others. I am having more and more conversations with leaders who are starting to ask themselves and their teams fundamental questions about how and why things are done.
In addition, the process of looking into our workflows as part of the use case discovery for AI, often reveals that we are not really acting the way that we claim to. That we are not actually following those best practices we supposedly put in place. That a lot of it has been theater for others, or worse yet - fooling ourselves.
A few exampes - from marketing, to engineering, to organizational structures
To make this more concrete, here are a few examples of rethinking common paradigms that came up at Tech Arena conversations I have had.
1. Is the marketing practice broken?
Marketing professionals have been optimizng their digital assets for search engines for more than 2 decades. Tools and best practices have been built. Content has been written in certain ways, and roles and teams with specific capabilities and skills have established.
But brand and product discovery is changing quickly. I touched on this in issue 1 of this newsletter, discussing the term Answer Engine Optimization. In a session by Trust Pilot, it was quite obvious that: a. No one really has a playbook for this now. b. It will keep changing quickly with the rapid evolution of AI models and products, and c. You need to write not just for humans anymore, but also for AI agents.
If you are in marketingm, you definitely need to have backbone and challenge many of your assumptions.
2. Will engineering ever be the same?
2025 has been the year of AI coding agents. It will only accelerate, as many of the leading players see AI powered software as an unlock for other use cases. This has been the focus of the latest model releases from OpenAI, Anthropic, Google, X.ai, and others. We have also seen tools such as Cursor, Claude Code, and Codex sweaping the stage. The role of the software engineer is becoming more that of an AI agent orchestrater and architect.
3. Rapid innovation, standards, and hackathons
I facilitated a panel at Tech Arena, titled How do you turn AI hype and one-off hackathons into real, repeatable innovation? In it, we were debating if innovation current practices and standards are still relevant in the age of AI, and what “truths” should we hold on to when driving innovation, and which structures, roles and responsibilites have to be reshaped. I must say - I don’t think we have good answers for these questions.
Your action step
Pick one established process in your organization - marketing, engineering, innovation, hiring, whatever feels most “settled.” Block 30 minutes with the people closest to it and ask two questions:
Are we actually doing this the way we claim?
If we started from scratch today, with AI as a given, would we design it the same way?
If the answer to either is no, you’ve found where backbone is needed.
Then flip the lens. Look at where your team is adopting AI, and ask:
Are we doing this because it genuinely solves our problem, or because everyone else is doing it?
Are we adding a chatbot because customers need one, or because it’s the default AI feature?
Are we rushing to automate a workflow that we don’t fully understand yet?
The same backbone that should push us to rethink old processes should also push us to question new ones. Sometimes having backbone means saying “not yet” or “not like this” - even when the rest of the industry is racing ahead.
#2 - Business Value with AI
Rethinking How We Design AI Experiences
For years, the design process for digital products has followed a familiar sequence: research users, create personas, map journeys, write problem statements, brainstorm solutions, wireframe, test, iterate. It’s been taught in schools, codified in methodologies like Design Thinking and the Google Ventures Design Sprint, and treated as a mark of professional rigor. If you followed the process, you were doing good work.
The problem is - it doesn’t consistently produce great work. And in the AI era, it’s increasingly producing the wrong work.
I’ve seen this too. Teams follow the steps faithfully. They produce beautiful journey maps, well-crafted personas, and perfectly formatted problem statements. They spend days and weeks on these artifacts. And then the actual product - the thing the user touches and feels - gets a fraction of that attention. The process becomes the deliverable instead of the experience.
Jenny Wen, Design Lead at Anthropic and formerly Director of Design at Figma, made this case sharply in a recent talk at the Hatch Conference. She described watching designers fill portfolios with 80% process artifacts and one screen of the actual product at the end. Her keen observation is that the user doesn’t care about your journey map. They care about whether the product makes them feel something.
This challenge is amplified with AI, where the technology changes what’s possible with every model release. The traditional sequence - understand the problem, then design the solution - assumes the technology is relatively stable. But AI isn’t stable. What was impossible three months ago is now a feature. You can’t map the full problem space for capabilities that didn’t exist when you started the project.
So what should replace the old process? Based on Jenny’s principles, my own client work, and what I’m seeing in the most successful AI products, here are five principles for creating AI experiences that actually deliver user and business value.
1. Start with the solution or the technology - not the problem
This feels counterintuitive, maybe even reckless. But consider how Claude Artifacts came to life at Anthropic. A researcher built a rough prototype that rendered Claude’s output as an interactive panel. No problem statement preceded it. A designer on the team saw it, recognized something was there, iterated on it, and they shipped it. It became one of the most influential AI interaction patterns of the past two years. As Jenny put it: “We didn’t know it was a problem worth solving until we actually saw the solution.”
This is something I apply in opportunity discovery workshops with clients. Rather than starting with journey maps, I guide teams to first understand what current AI capabilities actually are - what the models can do today in language, vision, reasoning, and tool use - and then match those capabilities to user and business problems. The technology illuminates problems you wouldn’t have identified otherwise.
2. Do less - skip steps and make up new ones
Jenny described running design sprints in three days instead of five, cutting the steps that weren’t producing value and spending most of the time on prototypes - because prototypes are what make people understand a concept. You could say she replaced Amazon’s press release exercise with imagining tweets - how would users react on social media to this feature? - because that felt more relevant to Figma’s audience.
The point isn’t that her shortcuts are universal. The point is that every project demands its own process. As she put it, you don’t know if you’re building a bookshelf or a hot dog - so how can the same set of IKEA instructions work every time?
3. Don’t trust the process - operate on intuition and conviction
This connects directly to the Have Backbone principle in Section 1. Jenny argues that intuition is not guessing - it’s the ability to make sound judgments quickly because you know the domain deeply. She builds her intuition by reading user feedback constantly, attending research sessions across the entire product, and watching usage dashboards regularly. That deep familiarity lets her make design decisions without needing to research or A/B test every one.
For leaders building AI products or features, the equivalent is building your own AI intuition through hands-on use - not just reading about what’s possible, but using the tools yourself, regularly, so you develop a feel for what works and what doesn’t.
4. Care ruthlessly about the details
After launching FigJam, Jenny’s team didn’t rush to add new features. They spent years iterating on the core experience - snapping, alignment, selection borders, colors, font sizing, shape overflow behavior, toolbar interactions. This long tail of quality refinement is what turned a functional product into one people loved.
AI products especially need this attention. The difference between an AI feature that gets abandoned after one try and one that becomes part of someone’s daily workflow often comes down to these details - how fast it responds, how clearly it shows its reasoning, how naturally it handles follow-up questions, how gracefully it recovers from mistakes. The fun part is that you can do all that much faster now. I completely redesigned the user flow for Castifai in a couple of days, based on feedback from early users.
5. Design to make people smile
FigJam’s stamps, emotes, and Cursor chat features - some of the most beloved parts of major products - didn’t come from a problem statement or a user journey. They came from a team that prototyped ideas in real code, showed up to people’s meetings, and watched their reactions. The features had usability issues. But people were smiling and laughing. That signal was more valuable than any research report.
When designing AI experiences, ask: where is the moment of delight? Not every feature needs it, but the ones that make people smile are the ones that get shared, remembered, and used again.
These principles don’t replace all research or strategy. They rebalance the equation - from process worship to outcome obsession, from following steps to building judgment.
For the full argument and the stories behind these principles watch Jenny Wen’s talk from Hatch Conference Berlin: “Don’t Trust the Process”.
#3 - AI Tool Spotlight: Shape of AI
A Pattern Library for Designing AI Experiences
In Section 2 I argued that the old design process isn’t producing great AI experiences - and that teams need new principles for building products people actually love. But when you sit down to design an AI feature, a practical question surfaces quickly: what patterns already exist? What are other teams doing to solve the blank canvas problem, build trust in AI outputs, or give users control over results? You don’t have to start from zero.
Shape of AI (shapeof.ai) is a free, open pattern library created by Emily Campbell - VP of Design at Hackers/Founders - that catalogs how AI is reshaping interaction design. It documents the specific UX patterns emerging across AI products, organized into categories that map directly to the design challenges teams face when building AI-powered experiences.
What it does
Shape of AI organizes AI interaction patterns into six categories. Each one addresses a different moment in the user’s relationship with AI:
Wayfinders help users get started. This is where many AI products fail - the blank text box with no guidance. The patterns here include suggestion chips, example galleries, templates, nudges, and randomize options. If you’ve ever opened an AI tool, stared at an empty prompt field, and didn’t know what to type, a wayfinder was missing.
Prompt Actions are the different things users can direct AI to do - summarize, expand, restyle, transform between formats, synthesize across sources, or regenerate. These patterns help teams think beyond the default chat interface and design specific, purposeful interactions.
Tuners let users refine and shape results. Attachments, filters, connectors to external data, model selection, modes, preset styles, and voice and tone controls all fall here. This is where user agency lives - the ability to say “not like that, like this” without starting over.
Governors maintain human oversight. This includes citations, action plans (showing the AI’s steps before it executes), verification prompts, branching, controls to pause or redirect mid-stream, cost estimates, and stream-of-thought displays. If you covered Human-in-the-Loop in Issue #2 of this newsletter — this is what it looks like in practice at the interface level.
Trust Builders help users believe the output. Caveats about model limitations, data ownership controls, disclosure labels, footprints that trace how the AI reached its answer, and incognito modes. These patterns are especially critical for enterprise use cases where accuracy and accountability matter.
Identifiers are the brand-level choices that make an AI experience recognizable - avatars, color systems, iconography, naming conventions, and personality definition. These seem cosmetic but they shape how users perceive and relate to the AI.
How I use it
When working with clients on AI features or when building my own AI products, I use Shape of AI in two ways. First, as a diagnostic tool. If an AI feature has low adoption, I walk through the categories and check which patterns are missing. Almost always, the issue is in Wayfinders (users don’t know how to start), Governors (users don’t trust the output), or Tuners (users can’t shape the result to their needs). The categories give you a structured way to find the gap without guessing.
Second, as a design reference in workshops. When a team is designing a new AI-powered experience - say a customer support assistant or an internal knowledge tool - I use the pattern categories as a checklist: How does the user get started? What actions can they take? How do they refine results? How do they stay in control? How do they know they can trust this? Going through these questions before building prevents the most common AI UX failures.
The site also includes a UI Library with real-world screenshots from products like Perplexity, Claude, Notion, Midjourney, Slack, and others - so you can see how these patterns are actually implemented, not just described.
What I like about it
Shape of AI fills a gap that didn’t have a good answer until recently. There are plenty of resources on how to build AI models, how to prompt them, and how to evaluate their performance. But there’s very little practical guidance on how to design the interaction layer - the part the user actually experiences. Emily’s library provides a shared vocabulary for teams that are designing AI features, which is especially valuable when designers, product managers, and engineers need to align on what “a good AI experience” actually means.
It’s also free, open (Creative Commons licensed).
For the full context behind why these patterns matter and how AI is reshaping interaction design, watch Emily Campbell’s talk from Hatch Conference 2024 (which surprisingely still apply): “The Shape of AI: How AI is Reshaping Interaction Design.” She traces the history from mainframes to algorithms to AI, and makes the case that AI can give agency back to users - if we design it that way. Her 10 Heuristics of AI - from “Purposeful and Needful” to “Identification and Honesty” - are a useful companion framework to the pattern library itself.
#4 - One New Term at a Time: MCP Apps
This Week’s Term: MCP Apps - an official extension to the Model Context Protocol (MCP) that allows AI tools to return interactive user interfaces - dashboards, forms, visualizations, workflows - directly inside the AI conversation, rather than just text.
In Issue #12, I introduced term MCP - the open standard that acts as a “USB-C port for AI,” letting AI models connect to external tools and data sources through a universal interface. Since then, MCP has grown rapidly. It was donated to the Agentic AI Foundation under the Linux Foundation, co-founded by Anthropic, Block and OpenAI. Over 500 public MCP servers exist today, and the protocol is supported across Claude, ChatGPT, VS Code, and many other platforms.
But MCP had a limitation. Tools could send data back to the AI, and the AI could summarize it in text - but users couldn’t interact with the results directly. If a tool returned a hundred rows of sales data, you’d have to prompt your way through it: “Show me last week’s numbers.” “Sort by revenue.” “What’s in row 47?” Every interaction required another message.
MCP Apps, announced in late January 2026, closes this gap. Now, when an AI tool returns results, it can also return an interactive interface - a chart you can filter, a form you can fill out, a map you can explore, a document you can review and annotate - all rendered inside the conversation. The AI stays in the loop and sees what you’re doing, but the interface handles what text alone can’t: sorting, filtering, clicking, dragging, and real-time updates.
For business leaders, this matters because it changes what AI-powered tools can feel like. Instead of a back-and-forth text exchange to explore data or configure a system, users get a familiar, visual experience - embedded right where they’re already working. It connects directly to this issue’s theme: the experience layer is becoming as important as the intelligence layer.
As Block’s Andrew Harvard put it in its announcement, the industry has been embedding assistants into individual apps, creating fragmented experiences. MCP inverts this - making apps pluggable components within agents. MCP Apps takes it further by bringing user interfaces into the agent experience itself. The same MCP App works across Claude, ChatGPT, VS Code, and Goose without writing client-specific code.
In the video below Den Delimarsky explains the importance of this move and how to get started.
Think Big Newsletter is created by Think Big Leaders -