Zachary Long Would Like a Word With the Rulebook

In July 2023, Zachary Long’s name appeared on an Air Force research award for something that sounds almost designed to repel a dinner-party conversation: “AI Assistance for Derivative Classification and Redaction.” The award was worth $1,073,424. Long was the principal investigator. Behind that uninviting title sat a question he would spend the next several years trying to answer: could software help people work out what they were allowed to do?

There is a certain comedy in founding a company to reduce paperwork and announcing yourself through a government award record. There is also a useful clue. Long was choosing a problem that required him to work inside the machinery. His future customers had obligations, classification guides, approval chains, and consequences for getting a decision wrong. A brisk speech about disruption would leave those obligations exactly where they were.

Today, Long is co-founder and CEO of ConductorAI. The company’s platform, Conduit, helps turn policies and institutional expertise into review workflows. His chosen subject is permission: who can grant it, which document governs it, and how long everyone else must wait.

A Brown degree, then the machinery of government

Long attended Brown University from 2009 to 2013 and earned a Sc.B. in Applied Mathematics and Economics. His later work at Palantir included data science and leading U.S. government financial-management work, with experience on the Army Vantage platform. It is a background in making institutional information useful, rather than merely making more of it available.

At Palantir, Long met Eric Schwartz on a Department of Justice project. They each spent seven years at the company. Their work subsequently diverged; Long concentrated further on defense. Benjamin Fichter, a college connection, helped originate the ConductorAI idea with him. Schwartz joined shortly afterward and became a co-founder and COO.

One frustration supplied the premise. Long wanted to put a meeting write-up in Slack, was refused, and asked to see the rule. Nobody could produce it. The small absurdity contained a business problem: a decision could be binding even when its explanation was inaccessible.

A missing explanation is a peculiar thing to build around. It does, however, give a founder a fairly exact brief. Make the rule findable. Make its application intelligible. Give the person waiting a route forward.

“Show me the document.”

Zachary Long, on the policy question behind ConductorAI

The first customers had every reason to be suspicious

Long’s early outreach involved sending cold LinkedIn messages to federal employees. Some recipients thought they were being tested for susceptibility to phishing. The misunderstanding became a joke with customers; it also suggested that the underlying problem was painful enough to get a stranger an answer.

The initial product idea concerned classification. The sharper need emerged when information had to move to a partner. That is when an abstract rule acquires a deadline and somebody needs a defensible decision. Long’s company adjusted around that moment of transfer.

By 2024, another Air Force research award had Long listed as principal investigator: $74,950 for the Machine Assisted Knowledge Exchange Repository, or MAKER. The project concerned Pacific Air Forces workflows and the difficulty of getting guidance drafted and approved quickly enough. Its description called for a tool capable of reading more than 500 policy documents in minutes.

The acronym is handsome enough. The underlying assignment is deliberately plain. People need to locate and interpret the instructions that constrain their work. Improving that task requires understanding why the instructions exist, how an office uses them, and where a review actually stops.

An answer with its working attached

Conduit’s architecture reflects that assignment. The platform searches structured and unstructured material, encodes policies and expert practices into sequences of decisions, and coordinates document reviews across teams. A user can work with office documents, images, video, and other material that refuses to arrive in one convenient format. The company also supports running decision steps through APIs.

Consider the kinds of work listed among its applications: sharing sensitive information with partners, reviewing export-controlled technical material, checking security classification, and finding evidence in large investigations. Each brings a different vocabulary. Each requires a reviewer to connect a proposed action to particular information and rules.

Long is building in a field where a fluent answer can still be an unfinished answer. The useful object is a recommendation someone can inspect and take responsibility for. A review has to survive the next question from the next person in the chain.

That makes the product’s interest in workflow more understandable. A conversation may help someone explore a problem. An approval process must also identify the steps, pass the work along, and preserve the reasoning. Bureaucracy has many faults; its appetite for follow-up questions is impressively consistent.

The second permission: getting bought

In April 2025, ConductorAI announced a $15 million Series A led by Lux Capital. The investor group included Altman Capital, Haystack Ventures, Sunflower Capital, Humba Ventures, Also Capital, Forward Deployed VC, and Abstract. The funding was intended to support expansion across government and other organizations wrestling with compliance.

For Long, the money backed a business proposition rooted in the employee’s working day. The platform could help locate the people, policies, and earlier approvals needed to complete a task. The company described Conduit as a research assistant and expert reviewer, adaptable to different customer environments and independent of a particular model.

Lux’s investment also connected Long and his fellow founders to an investor explicitly interested in the administrative work upstream of a mission. It is easy to admire equipment once it exists. The approval that lets information reach the right person is harder to photograph.

Funding supplied room to pursue the problem. Access to customers required additional work. Long’s company would need ways to enter established purchasing arrangements, work alongside existing systems, and fit into offices that could hardly suspend their obligations while a startup found its footing.

Working through the institutions

Booz Allen Ventures announced a strategic investment in June 2025. Its description of ConductorAI emphasized AI-assisted workflows with a person retaining judgment, including foreign disclosure and compliance. It also stressed making progress within existing government systems. The arrangement put Long’s young company alongside a business already deeply involved in federal work.

The following month brought a distribution partnership with Carahsoft. Conduit would be available to public-sector buyers through reseller partners and established contract vehicles, including NASA SEWP V and ITES-SW2. For a company selling improvements to approval processes, the route through procurement is part of the practical story.

Long described the mission in terms of improving government efficiency, particularly the approval and review work connecting agencies to companies and partners. There is an appealing symmetry here. His customers need a usable path through rules; his own company needs one too.

Neither a strategic investor nor a distributor supplies the final verdict on a product. They change the conditions under which it can be tried, purchased, and supported. Long’s progress can be read through these successive openings: research work, financing, institutional relationships, and a clearer route to a buyer’s desk.

Everybody goes on site

Long’s approach to customers also shows up in how ConductorAI talks about its staff. In a company update he shared, the team said every employee would go on site during the year and know users by name. The accompanying photograph looks like an ordinary gathering of colleagues. That ordinariness is useful. Government software is made by people who eventually have to meet the people using it.

Long has spoken about Palantir’s influence, the importance of being physically present, and the awkwardness of problems that many people experience without any single person owning them. When the company was roughly 90 days old, he and his co-founders were already flying to Pacific Air Forces, he recalled. He urged other founders to spend time beside users, even before every purpose of a visit was clear. Those are organizational questions as much as software questions. A problem can be familiar to an entire office and still lack a convenient budget line or a person empowered to change it.

His public manner has room for jokes. On a recruiting post featuring virtual-reality equipment, he suggested the team was playing chess. A separate basketball post advertised the office’s view of a Knicks parade and ended with a Celtics hashtag. Even a company concerned with permissions can tolerate a little provocation. Sports allegiance apparently requires no interagency review.

The question reaches Army Cyber Command

On August 27, 2026, ConductorAI announced a contract supporting U.S. Army Cyber Command through its Army Cyber Technology and Innovation Center, known as ArCTIC. The work concerned the command’s effort to use data, analytics, and AI in operational decisions. Information sharing was central to the announcement.

Long’s explanation focused on the government’s responsibility to approve the release of sensitive information to partners and the public. It is a useful statement of his position. The release still needs approval; he wants the review to move faster. The care required by the task remains part of the task.

The company described approval workflows that deployed organizations had seen move from weeks and days to hours and minutes. Those are reported customer experiences, rather than a universal stopwatch for government. The actual result depends on the workflow and its setting.

What the contract adds to Long’s story is a specific institutional setting for the idea. A founder can spend years describing a bottleneck. Work with a command gives that description a place where the software, the users, and the operational need have to meet.

Less waiting between knowing and doing

The next announcement arrived on October 6, 2026: an APFIT award supporting the Defense Intelligence Agency. ConductorAI said the funding would help deploy and scale Conduit. The selection marked DIA’s first award through the program and one of its early selections for a pure software solution.

By then, the company described its platform as FedRAMP High authorized and deployed across classification levels, with a Maine headquarters and offices in New York City and Arlington. Long, whose public profile places him in Brooklyn, was building a company whose work depends on moving between different institutional environments.

The direction remains consistent with the small problem that helped start it. Someone has information. Someone else needs it. Between them sit legitimate obligations, accumulated practices, and a review that must be completed by a person willing to stand behind the answer.

Long’s ambition is to shorten that interval. There is no need to imagine a government suddenly free of rules to understand the appeal. A reviewer finds the relevant policy. A colleague gets an answer. The work proceeds. For a founder who once asked to see the document, that is a satisfying place to keep looking.