"Where mission meets moonshot" was the theme of this year's Google Public Sector Summit in Washington, D.C., and it fit. The energy on the floor was about moving AI out of pilots and into real mission outcomes, with Gemini at the center of it.
Google Public Sector put a number on the challenge. In a survey of public sector leaders shared at the summit, 96.3% of agencies said they have deployed AI agents in some form, but only 9% have reached agency-wide adoption. The barriers Google named were legacy data silos and integration complexity.
That gap is exactly what I spent the summit talking about. Between conversations in our room, I sat down with Fed Gov Today to cover two things: a new product we launched with the Google Workspace team, and what agencies should be doing to their data right now if they want Gemini to deliver at scale.
The short version: the moonshot runs on data. Pointing AI at your entire data estate doesn't make it smarter. It makes it slower, more expensive, and more likely to be wrong.
One of the headline announcements at the summit was the Gemini agent, a single, universal agent for work that draws on an organization's own context to plan tasks, use tools, and return finished work. Google also introduced Gemini for government services, which connects Gemini to existing agency systems without migrating sensitive data.
Both are a big step forward for agencies, and both reward the same discipline: the quality of the output depends on the quality of the context underneath it.
As I put it in the interview, "it's significantly important to put the right data into the AI models, as opposed to just giving AI access to unstructured data: a lot of duplicates, a lot of data that's older than you perhaps anticipated, or maybe even is sensitive."
Three problems come with an unexamined data estate:
None of these improves the answer. All of them add to the bill.
We frequently hear from organizations that gave AI access to the entire corpus and trained it on the wrong things.
Take an auto manufacturer that wants AI to improve customer service and the experience at the shop. "Really you wanted to give it access to the last five years of repair history for that model, but instead you gave it access to 25 years of data," I explained. "Well, you find out that your model is not well trained, and there's opportunities for failure, or duplicative information, or nonsensical information, because you gave it access to too much data."
The same logic applies to an agency fleet, a case archive, or a program history. The right five years beats the whole archive.
Before scaling Gemini across an agency, look at the data estate through three filters:
Lightning IQ scans data in place, across on-premises and cloud environments, so agencies can apply those filters without first moving sensitive data. That matches the direction Google laid out at the summit. The goal is to find "how many data points or how many metrics are relevant to the AI initiative and really would help create the highest return on investment from that data," and then target it wherever it lives, "whether that's your on-prem environments or any cloud environment."
Fed Gov Today asked whether PII requires special steps. My answer: that's a question for whoever is running the models or building the agents. Our job is to give them the choice.
"Whether the answer is yes, you want to redact, or you want to remove and withhold data that contains PII, we are able to do that. Or if you simply want to flag data that contains sensitive information or PII."
And it goes beyond standard PII. Any pattern of letters and numbers can be identified: "If you're an auto manufacturer, a VIN number, for example, or if you are a European government, maybe a specific country identifier. We can identify all those patterns in the data and specifically produce those to an AI initiative or an information governance initiative, or withhold those from some training model."
That is how governance by design works in practice. The agency's policy decides. The data layer enforces it.
"Every investment in an AI initiative frequently concentrates around the tokens," I noted: "the amount of data going in, the tokens for processing, even the tokens required to export on the back end."
Google built cost controls directly into the new Gemini agent, which is the right instinct. The other half of that equation is upstream. Every unnecessary document is a cost with no return. Sending less, but better, data into Gemini supports better results and more predictable spending at the same time. The cheapest token is the one you never process.
At the summit, we launched Lightning Recall with the Google Workspace team for agencies and organizations carrying long or forever retention policies on archived Google Workspace users.
Lightning Recall lets organizations export those archived users to Google Cloud Storage without losing searchability. The data stays in Google Cloud, and teams get Vault-like search access to exported users so they can still fulfill legal and compliance requests.
The interviewer picked up on something important: in many cases this isn't only about keeping search. It's about gaining it. "It's a little bit like taking a Gmail search and then going to a desktop search," I said. "There's additional horsepower that we can add." For an agency with 100,000 users, "our software is not limited. It could search across all 100,000 exported users."
Why work with Google on this? Because, as I told Fed Gov Today, the Google Workspace team is "just as committed to making AI available to public sector organizations as they are to the private sector." They knew our ability to "search and identify petabytes' worth of data in hours or days instead of weeks and months," and saw how it could extend what Workspace customers already rely on.
Archived-user search and Gemini readiness depend on the exact same capability: knowing what data you have and quickly isolating the records that actually matter. That data foundation is what bridges the gap between an isolated pilot and agency-wide value.
Google's survey highlighted the distance between deploying an individual agent and achieving agency-wide adoption. Solving the underlying data problem is how agencies cross that divide. Lightning IQ has a long history of working with organizations on massive data volumes for information governance, cyber investigations, and AI readiness. The advice remains the same in every scenario: examine your data estate early, identify the years, subjects, and sensitivity categories that matter, and give Gemini data you can trust.
That is where mission truly meets moonshot.
Ready to prepare your data estate for Gemini? Learn more about Lightning Recall or talk to the Lightning IQ team. Lightning IQ is available now through Carahsoft for public sector buyers and soon via the Google Cloud Marketplace.
Read more:
Lightning Recall is a Lightning IQ solution, launched for all Google Workspace customers at the Google Public Sector Summit 2026, for organizations that retain data from archived Google Workspace users. It exports archived-user data to Google Cloud Storage while providing fast search across those records, ensuring legal and compliance teams can fulfill requests efficiently without Archive User Workspace license overhead.
Lightning Recall is built for any Google Workspace customers with long or permanent retention policies or []a large population of archived users. This includes federal, state, and local agencies, higher education institutions, and enterprises. The primary teams who use it are CIOs, IT and Workspace administrators, and eDiscovery, legal, and compliance personnel.
Yes. Lightning Recall exports archived-user data to Google Cloud Storage and provides dedicated search across every exported user, with no limit on user count. Legal, records, and compliance teams can continue to fulfill audits and public records requests after data moves off Archive User Workspace licenses.
Start by filtering data across three categories - relevance, date, and sensitivity - before connecting it to an AI workflow. Eliminate duplicate records and superseded policies, and determine whether PII should be flagged, redacted, or excluded entirely. Lightning IQ scans data in place across on-premises and multi-cloud environments, enabling agencies to enforce these guardrails without migrating sensitive data first.
Data readiness is the foundational requirement for any AI, governance, or migration initiative: knowing what data exists before models query it. Most agent initiatives stall because they assume clean, curated context is already in place. Establishing data readiness means scanning and classifying information where it lives - by relevance, date, and sensitivity - so Gemini operates on authoritative, cost-effective, and compliant context from day one.