You're pledging to donate if the project hits its minimum goal and gets approved. If not, your funds will be returned.
In grant and grant-adjacent applications right now, bullet points are the single biggest visual tell of AI drafting. Everyone uses LLMs to generate three to five tidy, bolded bullets per section. It feels manufactured, homogenous, and instantly lowers a reviewer's guard into "skim mode."
Your voice naturally works through cadence, rhythm, and paragraph breaks. Prose with short, declarative sentences carries a quiet, stubborn weight that bullet lists completely destroy. It sounds like a human sitting in a room, thinking clearly, and writing deliberately.
Here is the entire piece with zero bullets, running entirely on short paragraphs, focused phrases, and clean section breaks:
PrismEthics Workbench is a model-independent continuity and governance layer for serious human-AI work.
AI is remarkably capable now. The way we work with it is still scattered.
A serious project can live across ChatGPT, Claude, GitHub, browser tabs, notes, documents, and old conversations. Each tool does its job. The problem is what happens to the work between them.
I ran into this myself.
I would switch models and need something I had worked out earlier. I knew I had done the work. I just could not remember where it was.
Then there was the larger problem. Even when I found something, the surrounding context was often gone. A source had been separated from the claim it supported. A decision was still there, but not the reasoning behind it. An important question had disappeared into an old conversation.
AI creates another layer of uncertainty. A model can tell you that something was checked. That is not the same as showing that it was checked.
PrismEthics Workbench starts from a different idea: the Work Object.
A Work Object is simply the thing being worked on. It has a history. It accumulates evidence, decisions, questions, tasks, and finished work. The Workbench keeps those things together instead of scattering them across separate tools.
The goal is not another chatbot. It is a durable record of the work itself.
That record should remain intact when the AI changes. One model can handle coding. Another can work through a difficult argument. Another can help with critique or writing. The project stays with you.
The models may change. The work should not have to start over.
The architecture is already fairly developed. The next step is making it useful to people besides me.
The first goal is a practical Work Object. A user should be able to see what is known, what is assumed, what has been decided, and what still needs to be worked out. The supporting evidence should stay attached.
Next comes multi-model continuity. I want two or more AI providers to work from the same project. Switching models should not mean reconstructing the project from scratch.
Then there is governance. AI output needs different levels of trust. A suggestion is not a finding. A finding is not a verified result. And a model's response does not become part of the project merely because it sounds convincing. That distinction needs to live in the system itself.
Verification is the next piece. I am building Python-based checks and versioned records so the Workbench can show what actually happened rather than relying on the model's description of what happened. If the model says a test passed, the system should be able to ask one simple question: Where is the test result?
The last major goal is cognitive navigation. I have developed a library of 578 cognitive frames. Each frame is a question that changes what the AI looks at before it acts. One might expose a missing piece of evidence. Another might test an assumption. Another might reveal what the current way of looking at the problem leaves out.
The system also tracks obscures. An obscure is something important that has not been adequately brought into view: an evidence gap, a hidden dependency, a contradiction, or an assumption nobody has examined. A polished answer should not make those things disappear.
The intended workflow is straightforward: Work Object, to task, to frame, to plan, to bounded AI action, to artifact or evidence, to verification, to an authorized state update.
I expect that process to change once people use it. That is part of what this stage is for.
I am requesting $12,500 for about six weeks of focused development and pilot preparation.
Seven thousand dollars would support my development time. That work includes building the product, changing the architecture where needed, connecting it to other tools, debugging, testing, and documenting it well enough for someone else to use.
Two thousand dollars would cover AI model and coding-tool costs across Codex, Sol, and other models throughout development and testing.
One thousand dollars would cover technical services: hosting, automated testing, monitoring, backups, GitHub, security tools, and domains.
The remaining twenty-five hundred dollars would support testing, evaluation, user interviews, and pilot setup.
A smaller prototype could be built for $7,500. I could probably get a basic version working in four weeks. I am asking for $12,500 because I want to go far enough to test the actual idea, not just demonstrate it. That means multiple models, the full continuity layer, the frame and obscures systems, real testing, and real users.
At $15,000, I would have room for stronger security and recovery work, more time with early users, broader cross-model testing, and deeper evaluation. The target remains $12,500.
I want to speak with somewhere around 20 to 30 potential users.
They are people whose work continues long enough for context to become a problem. Their projects are not finished in one chat. I am especially interested in people doing research, software, long-form writing, policy work, analysis, or complicated project planning.
The purpose is not to sell them on PrismEthics. I want to find out where their current process breaks. What gets lost? What has to be reconstructed? Where does trust become difficult?
From those conversations, I want to find five to eight early users willing to bring real work into the Workbench. They would not be testing a canned demo. They would be using it on something they actually care about.
We would start by looking at their current workflow. Then we would decide what should change and how we would know whether it helped. Data access would be defined up front, including which AI providers could see which material.
At the end, I want a very simple answer: Keep using it. Pay for it. Expand it. Or stop.
Any of those answers is useful.
Early users would receive access, onboarding help, and a direct role in shaping the product. The pilot would be free or discounted. They would retain full control of their data, and nothing would be published without permission.
I am Eamon Montgomery, the founder and architect of PrismEthics.
My background is unusual for this kind of project. I am a special education teacher. I studied philosophy and cultural anthropology.
For years, my work has involved complicated human problems where the obvious answer is often not good enough. That eventually led me toward reasoning, conceptual change, institutions, ethics, and decision-making under uncertainty.
Then I began working seriously with AI.
I used different models for research, writing, coding, analysis, argument testing, and longer projects. Over time, AI stopped looking like a writing tool. It became part of the environment in which I was actually doing the work.
That was when the continuity problem became obvious. I started building a solution.
The work now includes the Temporal Persistence Lattice, the PrismEthics frame system, frame-routing logic, the Work Object architecture, obscures tracking, session continuity, and Python-based governance checks. The frame library contains 578 cognitive frames.
I have used the system in my own research, writing, institutional analysis, forecasting, and software development. I have also written publicly through PrismEthics about ethics, institutions, political economy, and AI-supported reasoning.
I built the initial system myself, with advanced AI models helping with coding, research, critique, testing, and other parts of the development process.
There is no large team behind it.
At this point, the intellectual and architectural core is mine. The next phase needs other people. I need users who will push against the system, technical collaborators who can make it better, and independent people willing to tell me when an idea does not hold up.
I think the problem is real. I am not sure the product is right.
People may decide they do not need another layer around their AI tools. The major AI companies may add enough project memory and continuity features to make a separate system unnecessary.
The frame system could also prove too complicated. It may help me because I know why it exists and still feel opaque to someone encountering it for the first time.
Verification has a limit, too. A system can show that a process happened without proving that the reasoning was correct. I want that distinction to remain clear.
And the market is moving quickly. Some of these capabilities could simply become standard features.
The actual use case may also narrow once people start using the product. The strongest value could turn out to be continuity. Or verification. Or provenance. I do not need to know that in advance. I need to find out.
That is why I want real users and a defined period of testing. If ordinary AI tools already solve the problem, I need to discover that. If the governance layer does not matter, I need that information too.
Failure would still leave me with something valuable: a tested system, evidence about the Work Object model, a better understanding of what people actually struggle with, and a much clearer sense of what should be simplified or abandoned.
I do not want to spend years defending an elaborate theory against reality. The point of this funding is to put the thing in contact with reality.
It could also become the beginning of a new career.
I have raised $0 in the last 12 months.
The research, architecture, writing, and initial development have all been funded independently.
There has been no outside investment and no previous grant funding.
This would be the first external funding for PrismEthics Workbench.