With businesses losing $37B annually to unproductive meetings, the explosion of AI meeting assistants promised a solution but created new problems: privacy risks, generic outputs, and inevitable price hikes. We're introducing a fundamentally different approach: by processing everything locally on Apple Silicon, Slipbox builds a rich understanding of your work context that improves with use, while keeping your data private and costs predictable - transforming meeting time from a cost center into a knowledge asset.
In 2023, Shopify implemented a "calendar purge" to cancel all recurring meetings and added a cost estimate to each meeting invite. The impact was immediate and measurable. Employees spent 33% less time in meetings, which led to a 25% increase in shipping speed. These operational improvements appeared to drive significant business results, as the company reported a 26% revenue increase in 2023 compared to the previous year. This wasn't just another productivity experiment - it was a clear signal that the way we think about meetings needs to change.
Figure 1. Shopify's calendar purge led and meeting cost estimate
The success of Shopify's calendar purge highlights a painful truth, while meetings are essential for collaboration, our current approach is broken. The average professional spends 10 hours weekly in meetings and preparation, with senior managers logging up to 23 hours - nearly three full workdays, resulting in a potential "wasted" annual investment of more than $25,000 per employee. Despite this massive time investment, businesses lose over $37 billion annually to unproductive meetings in the US alone.
The Meeting Assistant Gold Rush
This growing awareness of meeting costs has driven rapid innovation in meeting assistance tools and AI note-takers. The shift to remote work accelerated this trend, with dozens of new solutions emerging to help teams collaborate more efficiently. However, many of these tools focus on symptoms rather than underlying causes - adding features and automation while the fundamental challenges of effective meetings remain unsolved. The result is often increased complexity without proportional improvements in meeting effectiveness.
Figure 2. Meeting assistant companies founded since 2015 reported on Trxn. Note that this doesn’t even include all the players, some are categorized differently.
However, despite this explosion of sophisticated tools, three fundamental challenges remain unsolved:
- The Economics Problem: As AI costs stabilize rather than reaching zero, how can meeting assistants build sustainable business models without the dreaded "pricing rug pull"?
- The Personalization Gap: While meetings contain treasure troves of personal context and insights, most tools generate generic outputs that ignore individual working styles, preferences, and relationships.
- The Privacy Paradox: How do we balance the need for rich context and personalization with growing privacy concerns?
These challenges aren't just theoretical - they're reshaping how meeting tools are built, priced, and deployed. Let's start with the economics: why do meeting assistants struggle to build sustainable business models, and what does this mean for users?
AI Will Get Cheaper but Spend Will Grow
Figure 3. Historical price of computer memory and storage - Our World in Data
This is a classic example of Jevons paradox - while AI costs are decreasing, total spending will keep growing as usage expands - just as CPU, memory, and storage costs followed the same pattern, where decades of price reduction led to higher consumption rather than lower total costs. The cloud storage market tells this story clearly, it's projected to grow from $117B to $490B by 2033 (figure 4). Companies are still paying millions every year for EC2 instances, EBS volumes and S3 storage. Similarly, relying solely on cloud-based AI models for end-user applications would be like doing all your video editing or compiling in the cloud, instead of leveraging the processing power of your local machine.
Figure 4. Cloud Storage Market Share, Size, Trends and Growth 2033
As basic features like transcription and summaries become commoditized, costs still scale directly with usage. Companies may limit features, raise prices, or operate at unsustainable margins. Even as individual components get cheaper, total spend increases because lower prices drive higher usage.
The traditional playbook of "grow at all costs" and figure out economics later - popularized by companies like Amazon, Airbnb, and Uber - doesn't work for meeting assistance. Users are becoming increasingly wary of what they call the "pricing rug pull" - where companies offer value at a discount to gain market share, then increase prices once users are locked in. With transcription and basic AI features now becoming commoditized across the industry, companies can't rely on discounted pricing alone to build sustainable businesses.
Figure 5. X thread on startup pricing and "rug pulls"
These economic constraints create a vicious cycle that's reshaping the entire meeting assistant landscape. While businesses lose $37B annually to inefficient meetings, solving this problem isn't as straightforward as simply paying for better tools. The economics hit hard at every level - individuals face mounting costs as their usage grows, small and medium businesses struggle to balance comprehensive coverage with budget constraints, and even large enterprises have to justify growing AI expenses.
The challenge is equally daunting for meeting assistant providers themselves. The traditional SaaS playbook promised high margins, but AI has changed the game. In our previous blog, we discussed how cloud-based transcription services alone eat up 30-40% of margins - before even considering more advanced AI features. Companies are left with impossible choices: limit features, hike prices, or burn through cash with unsustainable unit economics.
This creates a vicious cycle: as companies try to control costs, they're forced to process meetings in increasingly generic ways. This leads us to our second challenge: the personalization gap. When every API call costs more than just the underlying compute, how can meeting assistants deliver truly personalized experiences?
The Power of Personal Context
Meetings are inherently personal, they are a gold mine of personal information and context. Every conversation reveals what matters to participants - their priorities, challenges, working styles, and relationships. This context is invaluable, yet most meeting platforms treat it as disposable data, either summarizing it generically or dumping it into another tool like Slack or Salesforce.
Think about your last few meetings. You might have discussed project timelines with team members who prefer detailed, step-by-step plans, or brainstormed with colleagues who thrive on high-level strategic discussions. Maybe, you've noticed how some stakeholders want detailed technical specifications while others respond better to visual presentations. Each interaction builds a rich tapestry of context about how people work and what they care about. Current meeting assistants miss this opportunity entirely, treating every meeting - and every user - the same way, generating generic summaries and insights that could apply to anyone. Even platforms that claim to offer "personalization" simply let users create templates or customize outputs. It's like having a personal assistant who never learns your preferences or work style - they just follow the same script every time.
Figure 6. Targeted In meeting insights delivered to you, personalized for you
We built Slipbox differently, inspired by the Zettelkatten method - a note-taking system designed to create connections between ideas. Instead of treating meetings as isolated events, Slipbox automatically breaks conversations into interconnected "slips" - atomic pieces of context about people, projects, and preferences. These slips form a living knowledge graph that grows more valuable over time. And it doesn't stop at meetings - you can easily import your own documents, notes, and research materials, allowing Slipbox to discover connections across all your knowledge sources. Whether it's meeting transcripts, PDF documents, or personal notes, everything gets processed locally to build a rich, interconnected web of insights. By processing this information locally, we can provide deeper personalization with less economic constraints of cloud-only solutions.
Figure 7. Slipbox personalized chat responses example, asking about feedback from our customers on our redesign
While this rich personal context enables powerful personalization, it also highlights our third major challenge: protecting sensitive information at scale. The more personalized and context-aware a system becomes, the more sensitive information it processes.
Privacy Matters
Meeting transcripts contain more than just conversation records - they capture sensitive company strategies, performance reviews, project postmortems, personal feedback sessions, and even casual team conversations about weekend plans. These discussions demand careful protection not just from external access, but often from different teams within the same organization.
This need for granular privacy control extends beyond traditional security measures. While security certifications like SOC 2 compliance are important, they address a different challenge than personal privacy. The data backs this up - when we interviewed users, 20% proactively asked about data privacy, and over 70% of FAQ clicks were related to data storage questions, with "Where is my data stored?" accounting for 34% of the clicks. This isn't surprising - nearly every user we talked to had personal notes they didn't want to share with others, even within their employer.
Traditional cloud-based processing makes this problem worse. It forces companies and users to store and process everything centrally in the vendor's cloud or jump through hoops to deploy on premise. This creates unnecessary privacy risks and makes it harder to leverage personal context effectively. For individual users, it means your sensitive meeting data is being stored by third-party vendors with little control over how it's used.
But there's an interesting dynamic we've observed from our early users - running AI locally doesn't just protect privacy, it enables deeper personalization through a virtuous cycle. Our users have told us they feel more comfortable sharing sensitive personal context with a system that runs on their own device. This richer personal context, in turn, enables more nuanced and personalized insights that wouldn't be possible with cloud-only solutions where users might hesitate to share sensitive information. As one user put it, it's a powerful feedback loop: local processing builds trust, trust encourages sharing more context, and more context enables better personalization.
We took a different approach with Slipbox. Your transcripts and chats stay on your device - and with recent advances in Small Language Models, your Mac can do a lot more than you might think. We can extract insights directly on your device, and these capabilities are growing rapidly as models get better and more efficient (see our previous blog). When we do need to process data on our servers, we don't store it - it's processed and discarded. Think of our server as a lightweight coordinator rather than a data warehouse. Your sensitive data stays within your control, only temporarily leaving your device when needed. We're also working on end-to-end encrypted backups to make this even more secure.
Figure 8. High level overview of Slipbox's architecture
This privacy-first architecture isn't just about security - it's the foundation for a fundamentally different kind of workspace.
Building Your Personal AI Workspace
The meeting assistant landscape is at a crossroads. The traditional cloud-first approach has created an unsustainable triangle of rising costs, generic experiences, and privacy concerns. By rethinking this foundation, we're showing there's a better way forward. We've intentionally started with Mac devices with Apple Silicon - though you'll get the best experience with M2 chips. This focused approach lets us tackle all three fundamental challenges head-on: sustainable economics through reduced cloud dependency, deep personalization through local knowledge graphs, and true privacy by keeping your data under your control.
Figure 9. The World’s Smallest AI Supercomputer | NVIDIA Project DIGITS
The future of AI computing is rapidly evolving. Projects like NVIDIA DIGITs demonstrate how powerful local AI processing is becoming, enabling a $3000 device to run sophisticated AI models that would traditionally require massive cloud infrastructure. This shift isn't just about technology - it's about reimagining the relationship between our personal data, our devices, and our digital workspaces.
As a two-month-old product, we're tackling the $37 billion problem of ineffective meetings head-on. While companies like Shopify showed that rethinking meetings can drive massive improvements in productivity, we're giving individuals the tools to make every meeting count. Local AI processing enables real-time, personalized meeting assistance that transforms how teams collaborate. We're building the foundation for truly effective meetings that respect your privacy, leverage your device's power, and scale sustainably with your needs. While we're still polishing some features, we're improving rapidly based on user feedback, which we share weekly in our public changelog.
If you have a Mac with Apple Silicon, you're already holding the key to a more powerful, private, and personal AI workspace. Download Slipbox and join us in building the future of meetings - where your conversations become insights, your privacy remains intact, and your workspace actually works for you. And because we process most things locally, you won't face unexpected price hikes or usage limits as you scale - no pricing rug pull here. Yes, we're still polishing some features, but the core experience is already transformative. Windows and mobile users, stay tuned.