It's Time to Assume AI Can Do Our Jobs

Over the last several weeks, concerns about AI have reached a point of public salience. First, uproar over data centers disrupting communities across the country, then hacks like the Hugging Face incident, and most recently a spate of AI safety experts resigning from major labs and sounding the alarm about imminent catastrophe.

As the leader of my company, today I sit here with the assignment of touting a new product we're really proud of, an AI prospecting service we believe is the best grant-finding experience on the internet. Yet I am conflicted about sharing this news, given the wider context of concerns over the speed and direction of AI development, which I share.

This latest build has increased my confidence that a vast number of human jobs can be disrupted or displaced by digital intelligence. The models have gotten way smarter, the people deploying them (people like me) have gotten better at it, and the manual workflows all around us are being deconstructed and handed over to agents.

As a former grant pro, I found the work of building our AI prospector fairly straightforward. Look at the steps involved in finding grant opportunities, break the steps into tasks and sub-tasks, create specialized AI steps for each, test and refine.

For example, human grant prospectors often begin by searching a grants database. We realized agents can read database information 100 times faster than humans, so we created database search tools meant specifically for agents to use. In the time it takes for me to review one funder profile, our AI prospector can analyze 100 or more.

Hundreds of billions, perhaps trillions, of dollars are being invested in processes like this for jobs across the economy. Everything from folding laundry to accounting to reading X-rays. People smarter than I, with a million times more resources, are creating ways to replace human roles with AI. Given what we've been able to do, I think many of those folks are going to succeed.

I chose to focus on grants in part because it was my background, but also because I had experienced the deep inequity of the grant funding ecosystem before I started my company. There is work in this sector, like the mind-numbing trawling through grants databases, that I believe should be handed over to machines. See also: adjusting character and word limits, recycling boilerplate, wrangling data, customizing budgets to funder preferences, etc.

We have complained about these things wasting our time, for years. Now, we have technology that can give us that time back. I have made the conscious choice to harness LLMs to replace and retire these burdensome tasks for good. But automation doesn't necessarily stop at the jobs we loathe.

I have tried to avoid the cliché, "AI won't take your job, someone using AI will." Instead, I have asked people to imagine how AI might replace their work. We no longer need to imagine. It is time for us all to assume that everything we do is plausibly replicable by AI. As a community, we should be deciding what makes sense to hand over, what we need to protect, and how we must shift our work to higher ground, so to speak, where we can build enduring human-centric roles anew.

If finding grant funding opportunities now takes a few minutes and costs maybe a couple bucks, what does that mean we should do with those savings and efficiencies? The opportunities and briefs AI generates still need the judgment and relationships held by people. I believe the vast majority of fundable organizations are under-granted, and most lack the capacity to stand up sustainable grant-seeking programs.

We have discussed the ethics and responsible use of AI for nearly four years. In that time, the sector has quietly moved to almost ubiquitous use in some form or other. To what end? What is the purpose of grant funding, and how can we anticipate the moment to serve the fundamental mission we are all a part of? Can we put these pieces together? Can we lower the cost and increase access to expertise and funding? I think so.