Revolutionizing browsing: inside Strawberry's AI-Powered workflow with Charles Maddock

Charles Maddock is the 24-year-old founder of Strawberry Browser, a Chromium-based standalone browser with a built-in team of AI companions designed to learn your workflows passively and automate your grunt work - without requiring you to write a single line of code. Charles started coding at 11, built his first AI agents in 2019 (an ecosystem simulator with neural-net microorganisms that won one of the world's largest international science competitions), co-founded a green tech company that was acquired, shipped a multiplayer video game, and did consulting for the Swedish government - all before 24. The GPT-4 API launch in early 2023 redirected everything: he started building browser-based web agents for developers, and that SDK evolved into Strawberry Browser.

Strawberry launched on Product Hunt and hit number one. The product is in closed beta, backed by a pre-seed round from a tier-one European investor and several US angels. The team is four people based in Stockholm, with plans to relocate to San Francisco. The vision is a browser where AI companions - named characters with distinct personalities - handle all the repetitive, copy-paste, data-wrangling work that currently interrupts flow state, while the user focuses on the work only they can do.

What Strawberry Browser Actually Does

Strawberry is built on Chromium - the same open-source foundation as Chrome - but ships as a fully standalone browser with its own branding, interface, and AI layer baked in. The core differentiator is a team of AI companions that live inside the browser permanently. They learn from watching you work (passively) or from explicit instruction (you record your screen and walk them through a workflow), and then automate the tasks you keep repeating - without any code, configuration, or third-party integration setup.

Because the companions operate inside the browser, they have native access to whatever is on screen - authenticated sessions, rendered page data, form fields, tabs - in a way that external agents cannot replicate without browser extensions or API access. They can click, scrape, fill, navigate, and export data across any website the user is logged into, including walled gardens like LinkedIn that have no public API. The current highest-value use case is structured data extraction: pulling information from a web page and writing it directly into a Google Sheet or CRM, eliminating the manual copy-paste loop entirely.

Companions, Not Agents: The UX Philosophy

Charles made a deliberate choice to call the AI entities “companions” rather than “agents.” Agents feels hard, technical, and distant. Companions - with names (Camilla for competitive research, Denise for data extraction, Linus for LinkedIn), personalities, and a collaborative dynamic - feel like your team. The intended experience is not a command-line interface with a chat box; it is more like a standup call with four faces in boxes, nodding their heads as you share your screen and explain what's driving you crazy.

The vision Charles describes for onboarding a companion to a new workflow: jump on a call, say “hi agents,” share your screen, do the task once while they watch and ask clarifying questions, and when they feel confident, end the call. From that point on, they handle it in background tabs while you do something else. No drag-and-drop workflow builders. No YAML configuration. Just showing an intern how you want something done. The companions can also collaborate with each other - “Hey Camilla, grab Denise and work on this” - routing tasks to the companion best suited to each step.

Micro-Friction Reduction and the Compounding Value of Flow

One of the most grounded product insights Charles offers is the value of resolving micro-frictions - the small, individually insignificant interruptions that add up to lost hours and broken concentration. The moment you stop to Google where to find an API key in a dashboard. The three seconds you wait for an answer that breaks your train of thought. The Monday morning bookmark cleanup you keep pushing to next week. None of these feel like product opportunities in isolation, but their cumulative drag on focus and productivity is real.

Strawberry's ambient presence in the browser means it can resolve these frictions without a mode switch. You ask the sidebar companion where the API key is; it points to the right link on the page. You mention to the browser that bookmarks you haven't touched in three months are cluttering your toolbar; a companion handles the cleanup. The goal is a full day of flow state where the friction never accumulates enough to pull you out - and that is the compounding value of having AI embedded in the place where you already work, rather than in a separate tab you have to remember to open.

Why the Browser Is the Right Layer for AI Companions

Most AI tools require a destination - a website you go to, an app you download, a tab you switch to. The browser is different: it is already where most knowledge work happens. By embedding AI companions directly into the browser rather than adding another destination, Strawberry removes the adoption friction that kills most productivity tools. The assistant is not something you have to remember to use; it is simply there, in the context where you already are.

The second structural advantage is authenticated access. Browser-based companions can operate within logged-in sessions - Gmail, LinkedIn, Salesforce, QuickBooks, any SaaS product - without requiring API credentials, OAuth integrations, or developer setup. If you can use the website, your companions can use it too. This is what makes the receipt-to-ledger, LinkedIn-to-CRM, and calendar-sync automations Charles describes achievable without any code: the browser is already authenticated, and the companion just needs to learn the workflow.

Data Privacy in a Browser AI Product

The obvious privacy question with an AI browser is: who sees the data? Charles is direct about the current state: because the most capable models need to be cloud-hosted, data processed through Strawberry's AI features flows through the model provider (OpenAI, Anthropic, or equivalent) under their standard terms - no user logs retained, no training on user data. Strawberry itself does not log activity.

The longer-term roadmap points toward local models. As open-source models like DeepSeek distills improve and hardware keeps up, users will be able to run capable models entirely on their own machines - nothing leaving the device. For tasks involving sensitive business data, this matters enormously. Charles frames it as a toggle: for most tasks, cloud models are fine and far more capable; for tasks where data sensitivity is high, local model support is the answer. The architecture is being built to support both.

Cursor as the Model: How the Best AI Dev Tool Changed Everything

Charles credits Cursor - the AI-native code editor - as the single biggest productivity multiplier on his team, estimating a 3x-to-4x improvement in output. He came to it skeptically (it felt like VS Code but without all his preferences), and is now unambiguous about the impact. His practical advice for developers new to Cursor: learn to write encapsulated, single-responsibility code. When each file has a clear, isolated purpose, Cursor can generate or modify it reliably, and if you need to rewrite it later, the scope is small. The failure mode for vibe coding is letting context sprawl across a codebase until the model loses track of what it is doing.

The other skill Charles recommends developing: learning to read the model as it writes. After enough use, you develop an intuition for whether the model's first few lines indicate it actually understood the prompt or whether it is confidently going in the wrong direction. That early-recognition skill - knowing when to stop, reframe the prompt, and restart - saves significant debugging time and is the difference between Cursor being a force multiplier and a source of subtle bugs.

Tools & Resources Mentioned

• Strawberry Browser - AI-native Chromium browser with built-in AI companions for workflow automation; in closed beta. Product Hunt number one. Pricing: free tier / $20 personal / $40 teams.
• Cursor - Charles's most-used tool after Strawberry; AI-native code editor. He attributes 3–4x team productivity gains to it.
• Chromium - Open-source browser foundation Strawberry is built on; same underlying engine as Chrome.
• DeepSeek (distilled models) - Cited as a near-future candidate for local model execution inside Strawberry for privacy-sensitive workflows.
• Apple App Clip equivalent vision - Charles describes a companion teaching flow modeled on a video call standup: share screen, demonstrate workflow, companions learn and then run it autonomously in background tabs.

Frameworks

FAQ