I have spent a good part of this year asking people across Democratic politics how their organizations are actually using AI. The sample is small, but the answer is usually the same. Individuals are trying things. Someone in digital drafts email with an AI tool. Someone in research runs memos through one. Someone in comms brainstorms scripts. Maybe the TV firm has put together an ad using AI images. Most of it happens person by person, often on personal accounts, and most of it amounts to a glorified search function with more detail.
What nobody described to me, in any of those conversations, was a strategy. Not one organization where AI adoption was a decision the organization had made, rather than a thing individuals were quietly doing.
That gap is going to close over the next year and into 2028, because the vendor pitches are already circulating and the pressure to “do something about AI” is real. And I can tell you now what the first formal stage will look like, because it is the same shape every time an organization formalizes scattered individual behavior. It will look like four pilots. Digital pilots an email-drafting tool. Organizing pilots script generation for canvassers. Research pilots a synthesis tool for opposition memos. Comms pilots video scripts. Four well-intentioned strategists. Four reasonable use cases.
And zero coherent pilots, because nobody in the building will be looking at the four together. The organization’s voice shaped by three different tools that were never told to produce the same voice. The donor file touched by two of those tools without either team knowing the other has access. The research memos and the canvasser scripts quietly diverging in how they characterize the opposition, because they draw on different source sets. The senior staff who would spot those problems too busy running their own pilots to notice. Sound familiar? It’s exactly how too many campaigns already run.
Campaigns themselves are mostly not in this picture, and that is worth pausing on. A campaign is full-on from the day it launches. Nobody inside one has six spare weeks to evaluate a tool against a workflow. The piloting happens in the ecosystem around campaigns, at the committees, the state parties, the advocacy shops, the vendors. Campaigns inherit whatever those organizations learn. Right now, mostly, they are inheriting nothing.
This is the part of AI adoption in Democratic politics that I think the industry conversation is mostly missing. Pilots do not fail because the tools are bad. Pilots fail because these organizations, at the operational level, are not coordinated entities by default. The middle of the org chart, which is where AI decisions actually land, is the same middle that struggles to coordinate everything else.
The pattern, when it arrives, will be easy to recognize. Multiple, simultaneous, mostly-unvetted AI pilots, running in parallel across teams, with overlapping use cases, with no shared standards, and with no central place where the organization is learning anything about what works. The embryonic version is already here, in the person-by-person adoption nobody is tracking at all.
The vendor pitch decks describe this differently. They describe it as the organization “adopting AI.” The view from the inside is that the organization is adopting four AIs, badly, and the cost of adopting them badly is roughly the cost of not adopting at all, plus the cash going to four contracts.
A pilot, properly understood, is not a tool. It is a structured way of learning whether a tool fits a workflow. The shape of a proper pilot is small. One named owner. One specific workflow the tool is being tested against. One measurement plan. One pre-written set of conditions under which the pilot is judged a success, a failure, or worth extending. One retrospective at the end where the team writes down what they learned about the tool, the workflow, and the team’s own ability to integrate AI.
Most of what is being called a pilot across the ecosystem right now does not have any of that. There is a tool. There is a person using the tool. There is, sometimes, a deadline. There is, almost always, a vague sense that “we are piloting AI.” The decision at the end will largely rest on the one user’s feelings about if the tool was helpful or not, without anything measurable behind it. That is not a pilot. That is a trial.
A trial produces output. A pilot produces learning. Trials are fine if you are running one or two of them in a coordinated way. The organization I described will be running four in parallel, with no coordination, learning nothing.
The deeper problem here is that AI is one of the few capabilities a campaign or committee can adopt where the coordination failure inside the organization is the failure of the technology itself. A new texting platform does not need every team in the building to align on what good texting backend operations looks like, because the texting team is the team that owns texting. AI is different. AI shows up in fundraising, in field, in research, in comms, in finance, in compliance, in candidate prep, in operations, all at once, and almost always through different tools and different teams. The thing that has to coordinate is the organization itself.
The irony is sharper than most senior staff realize. The use case where AI could produce the most leverage in campaign work is the coordination problem. Shared context across teams. A single source of truth on what the candidate said about an issue, when, in what venue. A way to make every team’s work visible to every other team without requiring a meeting. None of the four pilots I described are aimed at that use case. Nothing in any conversation I have had this year is aimed at that use case. The capability that would unlock the most is the capability nobody is even trying.
That is the part of the conversation I want senior leadership to take seriously, because it is the part where the standing posture inside these organizations is going to keep producing the same failure mode for the next twelve months or more.
The standing posture is: each team adopts AI for its own work, on its own timeline, with its own vendor selection. That posture treats AI as a productivity tool, like a faster calculator. It is roughly the wrong posture for the shape of the technology. AI is not a calculator. It is a tool whose primary leverage is in the connections between work, not in the speed of any one piece of work. Adopting it team by team is like buying email accounts that cannot send mail to each other and calling it an email system.
When an organization gets this right, and so far I have heard glimpses rather than examples, it will not be because it made one clever hire. I have made this argument before and I will keep making it: AI cannot be a token staffer bolted onto the org chart. The coordination has to run deeper than any single job description. It is a discipline that lives with senior leadership and shows up in how every team plans its work, and its whole purpose is to see all the pilots, all the tools, all the use cases at once and ask the questions no individual team is positioned to ask. Are these four pilots solving the same problem? Are these three tools touching the same data? Are we training the team on four interfaces when we could be training them on one? Is our voice being shaped consistently across these workflows, or is it drifting?
Somebody senior does have to be accountable for making sure those questions get asked, the way a campaign manager is accountable for the budget without being the only person who thinks about money. But the accountability is the floor, not the fix. The fix is an organization where every team expects to answer those questions before a pilot starts, and where the aggregate of AI across the building is somebody’s standing agenda item rather than nobody’s.
None of this is glamorous. It does not get a pitch deck. It is also the most leveraged capability a Democratic organization can build, and almost none have it. The familiar failure modes are all versions of making it smaller than it is. It gets folded into a digital director’s other responsibilities, where it competes with the digital director’s actual job and loses. It gets handed to a single “AI person” hired to make the problem go away, which is tokenism wearing a lanyard. Or it gets outsourced to a vendor, which is the equivalent of asking the calculator company to be your CFO.
There is a smaller version of this argument that I think every organization can run on right now, even without that hire. The smaller version is: before any new AI pilot starts, somebody senior asks four questions:
What is the problem this pilot is solving?
Who else in this organization is solving the same problem with another tool?
What data does this tool touch, and is anybody else here already touching that data?
Who is the named owner, and what does this pilot’s success or failure actually mean for our next decision?
If the answers to those questions take more than ten minutes to compile, the organization has a coordination problem that no amount of additional tooling is going to fix. The work is to fix the coordination problem first. The AI conversation is downstream of it.
The test, for any organization starting down this road, is whether it treats each new AI pilot as a discrete win or loss, or treats the aggregate of AI inside the organization as the actual thing to evaluate. Nobody has the coordination muscle yet. The ones that build it first will not be the ones with the most tools.
A pilot that produces output but does not teach the organization anything about coordination is, by that standard, a pilot that failed. It will be hard to see, because the dashboards will still show output. The failure shows up six months later, when the organization is paying for four tools, none of which talk to the others, and the senior team is still in the same coordination meetings about the same coordination problems, because the technology never absorbed any of them.
The organizations that come out of this cycle with real AI capability are not going to be the ones that ran the most pilots. They are going to be the ones that ran fewer pilots, in coordination with each other, and learned the kind of thing that makes the next pilot ten percent easier to run well. That compounds. The other version does not. And the campaigns those organizations serve, the ones with no time to pilot anything, will inherit the difference either way.
Fewer pilots, coordinated, with named owners, and an organization-wide view of what is actually happening across all of them. That is the boring answer. It is, as usual, the right one.
AI in Politics is authored by Jonathan Barnes, CEO of Authentic and co-founder of Quiller. He has been recognized as a Campaigns & Elections AI Leader of the Year and an AI Pioneer of the Year.
Jonathan’s writing focuses on how Democratic campaigns, progressive organizations, and political firms can move beyond the scattered AI experiments. He makes the case for building practical, human-reviewed systems that transform fundraising, communications, operations, and strategic decision-making.