For years, I was involved in developer advocacy.
I organised meetups, spoke at many of them, and was invited to conferences by people in my network. At one point, I was even a Microsoft MVP, partly as a result of all that developer advocacy work.
I had experience. I had technical topics to talk about. I had a network of people who knew me and my work.
But there was one thing I could never seem to get right: getting selected through a conference Call for Papers (CFP).
I tried. A lot.
I submitted proposals to conference after conference, and the result was almost always the same: rejection.
It was frustrating. I could be invited to speak at a conference by someone who knew my work, yet struggle to get selected when submitting a proposal through the official process.
Eventually, I started wondering whether the problem was the content of my talks or the way I presented them.
This year, I tried something different
Instead of starting with a conference CFP and trying to write a proposal from scratch, I started with something I was already comfortable doing: writing a blog post.
I would choose a topic I wanted to speak about and write an article about it. Then I would give an AI agent two things:
- My blog post.
- The website of the conference I wanted to submit a proposal to.
I would ask the agent to use my article as the foundation for a conference proposal, taking the event's audience, themes, and submission guidelines into account.
The idea was simple: I would focus on the technical content and the ideas I wanted to share, while AI would help me turn them into a proposal tailored to the conference.
And then something unexpected happened.
The results changed dramatically
For years, I had submitted proposals without getting selected.
With this new approach, I experienced a one-in-three acceptance rate.
That's a striking difference from my previous experience.
I can't prove that AI alone caused the improvement. I may have chosen different topics, targeted different conferences, or simply submitted proposals that were a better fit. But the change was significant enough to make me wonder what was really happening.
Had I finally found a better way to sell my ideas?
Or was something else going on?
The question I can't stop asking
There are two possible explanations that interest me.
Possibility 1: AI is better at selling my ideas than I am
Perhaps I was never particularly good at writing conference proposals.
That doesn't mean I wasn't good at presenting, explaining technical concepts, or engaging an audience. Those are different skills.
A CFP proposal is a piece of persuasive writing. It has to convince a reviewer that a particular talk deserves a limited slot in a conference programme. It needs to communicate the value of the session, identify its audience, explain what attendees will learn, and distinguish itself from other submissions.
Maybe I was good at the technical side of developer advocacy but not as good at selling the value of my talks on paper.
If so, AI is helping me communicate the value of my own ideas more effectively.
There's nothing inherently wrong with that. We already use editors, colleagues, mentors, and writing tools to improve how we communicate.
But it does raise an uncomfortable question: how many good talks have been rejected over the years because their authors weren't particularly good at writing CFP proposals?
How much technical talent have conferences missed because the people evaluating submissions couldn't see the value behind an imperfect abstract?
Possibility 2: Are the reviewers using AI too?
Here's where things get more interesting.
What if the improvement isn't just about AI helping me write better proposals? What if AI is also involved in evaluating them?
Some conference organisers may use LLMs to help review, summarise, categorise, or shortlist CFP submissions. Others may rely entirely on human reviewers, and many may use a mixture of approaches.
I don't know whether the conferences that accepted my proposals used AI in their selection processes.
But if they did, it raises a fascinating possibility.
Perhaps my AI-generated proposals are better at communicating with the people evaluating them. Or perhaps they are better at matching the language, structure, and criteria that an AI-assisted review process is designed to identify.
If an LLM helps me frame my proposal and another LLM helps evaluate it, have I improved my communication with the conference organisers, or have I learned to produce something that another AI system is more likely to recognise as a strong submission?
Those are not necessarily the same thing.
A proposal can be clear, well structured, and closely aligned with a conference's stated themes without the underlying talk necessarily being more original or valuable.
Conversely, an excellent talk might be overlooked because its proposal doesn't communicate its strengths effectively.
Are we getting better at communicating ideas, or just better at optimising the text that gets those ideas selected?
I don't have an answer. But I think it's a question worth asking.
Maybe the real problem was never my technical expertise
There's another possibility: the two explanations aren't mutually exclusive.
AI may be helping me write better proposals while also making my submissions more aligned with the selection criteria, regardless of whether a human or an AI-assisted process is involved.
Giving an agent both my article and the conference website provides useful context. Rather than asking it to produce a generic abstract, I'm asking it to connect my ideas to a specific audience and event.
That could lead to clearer titles, more focused descriptions, more explicit learning outcomes, and a stronger explanation of why the session belongs at that conference.
These are useful improvements whether the reviewer is a person, an LLM, or a combination of both.
The important distinction is that a better proposal doesn't automatically mean a better talk. It means the value of the proposed talk may be easier to understand.
And perhaps that was the missing piece all along.
How I'd approach this experiment now
If you're a developer advocate, engineer, technical leader, or aspiring conference speaker, here's the workflow I'd recommend experimenting with:
- Write the article first. Develop the technical ideas in your own words before worrying about selling them.
- Give AI the conference context. Include the conference website, CFP criteria, target audience, and submission requirements.
- Ask for a tailored proposal. Have the agent explain the talk's value to that specific audience, rather than simply summarising your article.
- Review everything yourself. Make sure the proposal is technically accurate, reflects what you'll actually present, and still sounds like you.
- Track the outcomes. Record submissions, acceptances, rejections, topics, and the approach used to create each proposal.
If I wanted to investigate the bigger question, I'd go further. I'd compare proposals written entirely by me with AI-assisted versions of the same ideas, ideally across comparable conferences and with enough submissions to draw more meaningful conclusions.
I'd also want to know how much AI is actually being used in CFP evaluation. Without that information, it's impossible to conclude that AI-assisted submissions are succeeding because they are being evaluated by AI.
My one-in-three result is an observation, not a controlled experiment.
But it's an observation that has changed how I approach conference submissions.
The question I'd like to leave you with
For years, I struggled to get my proposals accepted despite having experience speaking at meetups, organising communities, and contributing to developer advocacy.
Then I changed how I created my proposals, introduced AI into the process, and started seeing a very different result.
Maybe AI is simply helping me communicate my expertise better.
Maybe it is helping me optimise for a selection process that increasingly involves AI.
Maybe it's a bit of both.
I don't know yet.
But I find the question much more interesting than the simple conclusion that AI can write a better abstract.
If AI can help us write proposals that get selected, and AI is also helping evaluate proposals, what exactly are we optimising for: better talks, better communication, or better compatibility between AI systems?
I'd genuinely like to hear from conference organisers, CFP reviewers, and other speakers.
Are you using LLMs to write or refine your submissions? Are you using them to review submissions? And have you noticed a difference?
Because I suspect this is a conversation the developer community needs to have.
