In the world of artificial intelligence, the concepts of tokens and context windows are fundamental to how Large Language Models (LLMs) operate. As Dharmesh Shah explains in a post on his Agent AI website (which you should totally check out btw), tokens are the sequences of characters that LLMs use to process text, while context windows refer to the amount of tokens an LLM can consider at any one time when generating responses. Context windows are important because they determine how much information the AI can “understand” to generate meaningful outputs.
LLMs like ChatGPT use context windows to monitor the inputs (what we prompt it with) and the outputs (the text it generates). I remember when ChatGPT first came out for example, there was such a small limit to how much I could copy/paste into the chat window (user interface). Not only would it error out if it was too long, but even parts of the chat would get “forgotten” because the total token count exceeded the context window size. Even now with the most recent GPT 4o version, the AI can lose track of parts of the conversation or context you provide even if it doesn’t run out of tokens.
Regardless of tokens or processing power, the concept of context and AI is amazing.
If context is insufficient, then the quality of the output suffers.
Imagine you needed to work with a colleague to put together a KPI report but only had 5 seconds to give them context (i.e. what’s going on, who the report is for, what data points you’re looking for, the format needed, etc..) The report (output) will likely be incomplete and inaccurate, causing a bunch of assumptions and errors. This scenario perfectly demonstrates the limitations of context windows in LLMs. Just as the quality of the report depends on the context provided, the value and impact of an AI’s output are directly tied to the amount of context it can process.
Let’s take it out of the AI world for a minute, and focus on the human world.
Our work and communications depend heavily on the context we share and understand. Think about a team project for example - if everyone has the full context like the project goals, timelines, stakeholder expectations, resources, assets, documents, previous conversations, etc. then any discussions will be productive, decisions are well-informed, and everyone feels aligned and ready to move forward. But if key pieces of information are missing or misunderstood, the outcome can become cumbersome and chaotic, with team members making incorrect assumptions, missing deadlines, or even delivering something totally different than expected. Again, if the context is insufficient, then the quality of the output suffers.
Now let’s apply the concept of context to our computers or desktops.
Imagine you’re working on a project or initiative. That project probably has bits and pieces of its context scattered among various apps - a strategy doc in Microsoft Word, a spreadsheet full of data in Excel, a few emails, a couple of Slack messages, maybe a task in Trello, notes from a JIRA board, a ChatGPT window, and even some calendar invites for upcoming meetings about it. Each of these applications operates in isolation, with no shared context around your project. You’re the one who ends up manually transferring information between apps, prompting and re-prompting ChatGPT (especially when context windows are full), keeping details in your head, and constantly switching between different tools.
This fragmented context means that our outputs - whether reports, projects, tasks, or communications - suffer in quality. But in this scenario, it’s more than just the quality of output that’s affected. The burden of managing context falls on us too. So we spend inordinate amounts of time copy/pasting, manually linking, and hacking our days trying to ensure all our tools are aligned. This overwhelming task of managing context leaves us with little time and resources to focus, get actual work done, be creative and truly productive.
Now imagine if our computers held all that context for us, all the time?
Imagine they had contextual awareness.
If each of our apps knew the context of everything we’re working on, then AI agents or Gen AI tools like Chat GPT wouldn’t need exorbitant manual prompting - they’d already have the context they need to function at their best. The impact and efficiency of our work would be totally transformed, unlocking new levels of human capacity, creativity, and productivity. This is what Reframe is building with the Organized Work Environment (OWE).
When AI tools operate within Reframe’s OWE, they will have access to a rich, shared context that extends beyond the limits of traditional context windows and desktop GUIs. These AI tools would be able to draw from the entire contextual environment, leading to more accurate and relevant outputs, enhancing their ability to support complex tasks and provide meaningful insights.
With Reframe, individual contexts and perspectives are captured and shared, creating a comprehensive collective context. This new combined reality enables AI to access and leverage the larger, unified context, supporting not just individual tasks but also the collective needs and initiative of the entire team. This powerful integration addresses the ‘impoverished idealization’ in computing that was created from the simplification of only digitizing paper, providing a richer, more supportive digital work environment that enhances both individual and collective human capacity.
Game changer!
No matter how advanced these AI tools become, this future state would not be possible if the environment within which they exist (our current desktops) don’t fundamentally change. There will be a workday where our AI assistants not only understand our tasks but also the broader context of our projects. A day when they can provide insights, automate routine tasks, and support our workflow in ways that were previously impossible. This is the future that Reframe envisions and is building - a future where technology truly enhances human capability.
Reframe’s Organized Work Environment revolutionizes the way we work by integrating all our tools and information into a cohesive system. By extending the concept of context to an entire work environment, Reframe will enable humans and AI tools to operate with unprecedented contextual awareness.
This transformation not only enhances AI capabilities but also boosts human capacity and creativity. What would be possible then?
Want to experience this future of work? Sign up to get early access to Reframe’s Organized Work Environment.
