Five years ago, Google backed away from a Pentagon government contract because thousands of employees protested that its tech might be used for lethal drone targeting. Today, however, Silicon Valley has far fewer qualms about developing tech for the U.S. Department of Defense.

So said four investors — Trae Stephens of Founders Fund, Bilal Zuberi of Lux Capital, Raj Shah of Shield Capital and longtime In-Q-Tel president Steve Bowsher — speaking at a startup event for military veterans today in San Francisco. Said Shah of the shift in attitude that he has observed personally: “The number of companies, founders, and entrepreneurs interested in national security broadly — I’ve never seen it at this level.”

Bowsher argued that the “reluctance of Silicon Valley to work with the [Defense Department] and intel community” was always “overblown,” adding that across his 16 year with In-Q-Tel, which is the CIA’s venture fund, his team has met with roughly 1,000 companies each year and just “five to 10 have turned us down, saying they weren’t interested in working with the customers we represent.”

We’ll have more from the panel in TechCrunch+ but wanted to share parts of our conversation that centered on Things to Consider when selling to the U.S. government, given that founders with commercial customers may be thinking increasingly of trying to sell their products and applications to the U.S. military. (This is particularly true of AI and cybersecurity and automation startups.)

We talked with the investors, for example, about mission creep, meaning how a startup that begins working with the government can ensure it doesn’t wind up spending the bulk of its time catering to it, owing to new requests for this and that.

Here Trae Stephens — who also cofounded Anduril, the maker of autonomous weapons systems that has aggressively courted government agencies from its outset — said that this kind of gradual shift in objectives is “exactly what makes it hard to [cater to commercial customers and the government] at an early stage.”

He said that a “lot of the programs that [enable founders to] do early business with the Department of Defense requires some, like, DoD-ization of your product for that use case.”

Though In-Q-Tel backed Anduril early on (for which Stephens said he is thankful), he offered that many companies that take money from government, including through its Small Business Innovation Research (SBIR) program, “end up building all of these very specific workflow steps that take them away from the commercial businesses needed to make” the business truly work.

Stephens relatedly noted that very few outfits can afford to chase after the military early on as did Anduril, because it “takes so long to get into production with the DoD that you have to be able to raise, basically, an infinite amount of seed dollars; otherwise, the company is going to die.”

Relatedly, we asked how so-called dual-use companies deal with their intellectual property rights once they’ve begun selling to the government. For example, you can imagine a scenario in which a tech helps the NSA identify certain types of people who are making certain types of calls, tech that the government doesn’t want being released to adversaries. Is there a way to sort that out in advance, we wondered?

Here, there was no easy answer other than: get the right help, and do it as fast as possible.

Zuberi recounted one cautionary tale centered around a Lux portfolio company that he said is now a “unicorn” but that had to get over an unwitting entanglement first. Said Zuberi: “I have a company that received a $100,000 [National Science Foundation] grant. Two guys started it in my office. I didn’t think much of it; I thought it was nice to have on their resume. Then they started to do a Series B raise, and one of the [interested] firms does diligence on what other contracts [the team might] have, and there was a clause in that NSF grant that said, ‘Hey, if the government needs [what you’re building], we can use it.’ So we had to wait six months while we negotiated with [someone] at the NSF who didn’t care about it at all to get that right back.”

Zuberi said he “would have paid them double the amount of the grant just to make it go away, but they said ‘No, you can’t do this, we can’t go back.’ So you can run into problems.”

Again, we’ll have more from this discussion soon, including about AI in military applications; we learned a lot — hopefully you will, too.