Bring Your Own System Prompt

Beyond “Horseless Carriages”

Pete Koomen’s recent article, ”AI Horseless Carriages,” strikes a resonant chord. He articulates a widespread frustration: while AI models themselves are becoming incredibly powerful, many AI applications feel clunky, misaligned, and ultimately, unhelpful—like “engines bolted onto old carriage designs.” His call for AI applications that allow users to define and edit the “system prompt” to reflect their genuine voice and intent is not just insightful; it’s a critical signpost for the future of truly useful AI.

Koomen’s experience with Gmail’s AI being “un-Pete” is a universal one. It highlights a fundamental truth: without deep personal context and user-defined operational parameters, AI defaults to a generic, often unhelpful, “AI Slop.” The solution, as Koomen suggests, isn’t just about slightly better prompts for each interaction, but about a persistent, user-crafted “system prompt” that teaches the AI to act as a genuine extension of the user.

We are building this at Jaces

Jaces is building a personal data ecosystem that allows users to bring-their-own-ontology and data (which is akin to their system prompt). MCP/tool calling within chatbots have made this possible.

But what does it truly take to enable a user to craft and maintain such a powerful, personalized “system prompt”? We believe it requires two of our foundational pillars that go hand-in-hand: a personal data lake and a personal ontology.

Personal Data lake: Contextual System Prompt

Koomen notes that if Gmail had access to his twenty years of emails, it should be able to draft a better prompt for him. This hints at a core requirement: for an AI to understand and emulate you, it needs access to your data, your history, your context. A user-editable system prompt, however sophisticated, operates in a vacuum if it cannot draw from the rich tapestry of an individual’s actual life and communications.

This is where the concept of a personal data lake becomes essential. Imagine a private, secure, and user-owned repository where your diverse digital footprint – emails, messages, notes, calendar events, even biometric data or life logs from AR glasses – can be consolidated. This isn’t about surrendering more data to large corporations; it’s about individuals reclaiming ownership and agency over their own information, creating a comprehensive, private dataset.

This data lake becomes the experiential ground truth from which a personalized AI can learn your unique communication style, your relationships, your knowledge, and your patterns of life. It’s the raw material that allows an AI to move beyond generic responses to genuinely reflect how you would articulate something. Agents parse and query their way to understanding your tone and style.

We Need Data To Be Free

We need our whole life’s data to be free from the clutches of large corporations. They will use it against us to create vices and manipulative tendencies. This endeavour needs to be open-source and free. We need to be able to own our data.

Personal Ontology: Axiological System Prompt

While a personal data lake provides the “what” – the factual and behavioral context – an effective, user-defined “system prompt” also needs the “why” and “how.” This is where a Personal Ontology comes into play, allowing users to “Bring Your Own Ontology” (BYOO).

Koomen’s “Pete system prompt” is a brilliant example of defining desired AI behavior. A personal ontology takes this concept further, providing a structured way for individuals to articulate not just their communication style, but their core values, their goals (their telos), their priorities, their relationships, and even the inherent dialectics they navigate. It’s about defining the very framework through which the AI perceives, processes, and acts on your behalf.

If the personal data lake is the library of your life, the personal ontology is the librarian’s detailed notes on how to interpret and use that library in a way that aligns with your fundamental self. It transforms the AI from a mere mimic into an assistant that understands your intentions, respects your values, and actively works towards your defined objectives. This moves beyond simple preference settings into a deeper codification of individual identity and purpose for the AI.

We Need Purpose To Be Free

We need our whole life’s purpose to be free as well. I don’t trust a large corporation to define my life’s purpose, and neither should you. This would create for the mediocrity of the masses.

The Symbiosis: Data-Driven Context and Ontology-Driven Action

The real magic happens when these two components work in concert. Your personal data lake provides countless examples and the rich context for an AI to understand your world. Your personal ontology then guides the AI on how to use that understanding – how to prioritize, how to respond, what to create, what to ignore, and how to embody your desired persona.

Koomen’s vision of an AI email assistant that intelligently reads, categorizes, prioritizes, and drafts replies tailored to his needs is a perfect illustration. Such an assistant requires:

• Access to his email history (the data lake) to understand patterns and content.
• His “system prompt” or ontology to understand how to label, when to archive, what constitutes urgency for him, and how to draft a reply in his voice.

Let’s see an example:

Hi team,

Hope you're all having a productive week.

Just wanted to double-check if the final marketing assets for Project Carriage will be ready by COB this Friday. The launch is scheduled for next Wednesday, and we want to ensure all promotional channels are lined up well in advance.

Also, is there an updated one-pager on key talking points available?

Let me know if there are any blockers.

Best,
David

The Jaces example adds a personal brand voice, looked up the project and relevant files, added the user’s typical signature, and then drafted a reply with context. A vertical context approach (just emails) is not enough— it needs a horizontal context approach (project, files, brand voice, recent time spent on the project, etc.).

The Future is User-Forged AI

Pete Koomen is right: the “old world thinking” of developers as the sole authors of software behavior (and by extension, system prompts) is a constraint we must shed in the age of AI. The role of developers will shift towards creating the “agent builders” and “agent tools,” providing the platforms and capabilities for users to craft their own AI experiences.

The journey beyond “horseless carriages” in AI isn’t just about giving users a settings toggle for the system prompt. It’s about empowering them with true ownership of their digital context (personal data lakes) and providing sophisticated yet intuitive ways to impart their essence, values, and goals to their AI agents (personal ontologies). This is how we move from AI that performs generic tasks to AI that genuinely enhances individual agency, creativity, and focus, allowing us, as Koomen hopes, to “spend our time on things we do [love].”