From 100 Revisions to One Shift. Stefanie’s Journal Part I

I am currently going through the Grant Foundry Accelerator (GFA) as a client. I also coach clients through the programs. So I have an unusually complete view of what this process looks like from both sides, and I want to share that honestly, including the parts that are hard, the parts that surprised me, and what actually makes this program different from anything else I have seen in the grant writing space.

The program has four phases: Preparation, Play, Planning, and Performance. I will walk you through the first three, because that is as far as I have gotten with my current NIH R01. But before I do, I need to say something about how we’re using AI in GFA.

Most people I know are now using AI to support their writing. So, I want to be specific about how we use AI, because it is quite different.

Morgan is a computer scientist. And she has spent nearly twenty years developing frameworks for how researchers think about their science and their grants. So add these two competencies together, and something amazing comes out on the other end! The AI agents she builds are trained on the frameworks she developed for developing powerful proposals. Each client gets their own agent that starts learning your specific project from day one, accumulating everything you generate across phases: your research strengths, your field's landscape, your reviewer community, your big idea. It gets smarter about your project over time, in a way that a generic AI tool simply cannot replicate.

As coaches, we have our own separate agents. These are trained to help us assess client work at each phase, catch oversights that could become serious problems later in the process, and generate first drafts of phase outputs based on what clients have produced. I want to take a minute to highlight how different this is from the way we taught grant writing in the past and what other folks are doing: We use these AI agents to make sure the important things do not slip through the cracks and to show up to every coaching interaction with a next concrete step to work on, not just our vague impressions of the topic. 

This dissociates our work with clients, to a degree, from their ability to express their genius in spoken or written language. You probably know what I am talking about if you have ever listened to a talk and got so overwhelmed by the sheer amount of detail that you had a hard time following the red thread. We used to spend a lot of coaching time with clients just trying to understand and pull out the gold nuggets. 

Morgan’s AI agents help us do that much faster when we prepare for client calls, so that we can skip a lot of the back and forth. This is AI as infrastructure, built on deep expertise and deployed strategically so that human thinking, both the client's and the coach's, can move faster and go deeper than it otherwise could.

Preparation phase is about laying all the puzzle pieces on the table. Your research strengths, your potential projects, your funding landscape, and where the reviewer community's thinking currently lives. You work through a structured mind map with prompts, one at a time, building a rich database of information without needing to stress about making the right decisions upfront. You are just inputting data and doing targeted background research on your project, your funders, and your reviewers.

That database then feeds the AI agents, which generate draft strategy statements. But here is where the coach's job becomes critical. In the calls, I am not just listening — I am actively identifying where the client has real clarity and where the gaps are. Where do we need more information before we can arrive at a strong strategy statement? What is the client assuming rather than actually knowing? My coaching agent helps me assess the client's work systematically, flag potential oversights, and generate a draft strategy document that the client then corrects and deepens with their own technical knowledge, field pattern recognition, and scientific expertise.

The output is a strategy statement: a clear decision about what the project is and what it is not, where it will be submitted, and why the alignment is strong between the researcher's strengths, the project, and the reviewer community.

From my own experience as a client: I already had a clear project. What I did not have was clarity on which angle to use and whether to stay with my previous reviewer community or choose a different one. Preparation phase gave me the framework to answer those questions deliberately instead of defaulting to habit. It took about ten hours of homework and two coaching calls. The mind map looks intimidating at first, but once you are in it, the structure supports you. You do not have to figure out what to think about or worry if you are conveying your thoughts clearly. You just answer the next question.

The play phase is the hardest. I say that as someone who has been writing grants for over a decade.

It is called play because it is meant to be exploratory and creative. But creative and exploratory means it is not linear. You cannot put in ten hours and expect the output to be waiting for you at the end. And I want to be honest about what that felt like from the inside: it is uncomfortable. There is a particular kind of unease that comes from knowing you are supposed to be arriving somewhere and not quite being there yet — not because you are not working hard enough, but because the thinking is genuinely not done. As my writing coach, Louisa, helped me see that discomfort is actually diagnostic. When the words on the page feel slightly off, when your articulation of the field does not quite capture what you actually know, that friction is information. It means you are not done yet. You use it to push further rather than settling for something that is close enough but not quite right.

What you are working toward is the big idea: a few precise sentences that capture what is genuinely new, exciting, and timely about your project. Getting there requires being able to articulate three things cleanly: what the field currently believes — the old model — what does not fit or what is missing — the motivating problem — and how your project shifts things — the new model. Reviewers assess whether a proposal represents incremental progress or a genuine shift in how the field thinks. The play phase is how you make sure your answer to that question is yes, and how you learn to articulate why.

The client's AI agent is genuinely useful here. It can generate multiple different frames of the old model, the motivating problem, and the new model, helping you work in layers rather than staring at a blank page. But AI is verbose, and a significant part of the end work in the play phase is trimming: getting from a sprawling, conceptually rich articulation down to something short and precise that clearly conveys your Big Idea.

From the coach's side, my job is not to take away the discomfort. I cannot do that, and trying would be counterproductive. My job is to stay curious, pick up the interesting ideas that surface in the client's thinking, reflect them back, and ask: Does this belong in the old model? The new model? Does this go anywhere? We also bring in technical coaches — experienced scientists — to push on assumptions and close gaps the grant writer may not even realize they are still holding to evaluate the Big Idea. My coaching agent helps me to poke holes in the new model or approach. Better me than the reviewers!

Planning phase feels different. It starts to feel linear again. You can work through it systematically, and I have to say — I genuinely loved it.

The core work here is building out what Morgan calls defensive and offensive frames. Defensive frames are the reviewer concerns: what might they object to, why, and how do you preemptively address it. Most grant writers know how to do defensive framing. But here is what I noticed in myself: when you are on your third or fourth revision of a proposal that has already been through multiple review cycles, you can get trapped in a purely defensive mindset. You are so focused on the holes from the last round that you lose sight of what could actually get a reviewer excited enough to champion your proposal. I recognized this pattern in myself and it was uncomfortable to see clearly. Revision cycles have a way of slowly narrowing your thinking without you realizing it.

Offensive frames are about that excitement: what problems or challenges are holding the field back, what solutions would actually move things forward, what pieces of knowledge or technology could make a real difference. These are what make reviewers want to fund you, not just find no reason to reject you. There is a meaningful difference, and planning phase forces you to think through both dimensions deliberately.

These frames also have to be worked at multiple levels of specificity. At the broadest level, the question is whether your project matters to humanity. At the field level, whether it moves your research community forward. At the subfield level — for me, that means glial biology versus astrocyte biology specifically — you have to account for the fact that not all reviewers share your level of specialization. Depending on which study section you are targeting, you may need to frame the problem differently, do more educational work, or approach it from a different angle. Planning phase builds that analysis in from the start.

This is also where the AI does something that genuinely changed how I work. The client agent has been learning your project throughout Preparation and Play phase. By the time Planning phase rolls around, it knows your big idea, your old and new models, your reviewer community. It can now apply all of that to 1) help you identify defensive and offensive frames, and also to do some serious literature research. I spent one full day, 10 hours, on defensive frames for my current grant. My brain was completely fried at the end. But it was the productive kind of fried, because I was thinking, not searching. Questions that used to send my research team on weeks of literature search & analysis quests are now getting answered in half an hour of careful AI prompting. I am doing spot checks — verifying that the AI interpreted findings the way I would, assessing the rigor of specific studies, steering the search with my own expertise. This workflow accelerated the pace of actual thinking in a way I did not expect. I could ask more precise questions, learn faster what is and is not already known, and build a proposal that goes deeper because I actually had time to go deeper.

The output of this part of the Planning phase feeds into the Tower of Trust, where the proposal architecture starts to take shape. More on that in a future post.

If you want to follow her adventure, make sure to follow up with part II of Stefanie's journey.