Dragons hoard gold. AI knowledge is, for now at least, even better than gold. Early adopters are in an interesting position. Do we hoard it or do we spread the wealth?
A few years ago I was on a project built on a platform I'd never touched. Unfamiliar, unforgiving, the kind of thing where one wrong assumption costs you a day. We'd brought in an outside expert to help. He could plan. He struggled to build. Somewhere in the middle of that project it became clear that if it was going to ship, I was the one shipping it.
Becoming an expert was out of the question. I simply didn't have the time. But thankfully, ChatGPT was just released. I leaned on it the way you'd lean on a colleague who happened to know the one thing you didn't. Ask, verify, apply, move on. It wasn't magic. It was still my code, my judgment, my hands on the keyboard. But without it, that project doesn't make it out the door.
I didn't feel like I could be particularly public about why I was able to pull that rabbit out of that hat.
The cost nobody reimburses
I personally spend nearly $600 a month on various AI subscriptions and API fees so I can stay ahead of the curve. I know I'm not alone. There are countless early adopters who put in their own money and their own time to skill up in a technology that was obvious to us was going to change the way the entire world works.
It is wild that employers who were once hostile toward this shift are now simply expecting the people they employ to hand over everything they built on their own time and their own dime. There is no offer of additional compensation, no meaningful recognition.
This is wrong, almost laughable, and sadly, completely expected.
That's not a line in the sand. It's just true, and it's worth saying plainly: it's my knowledge. You don't automatically have a right to it. You can ask for it. And if you're me, if the ask is genuine, then the answer is "hell yes, where do we start?" But I also completely understand people who tell their employer to pound sand.
Loud no, quiet yes
I'll level with you, we're not actually dragons. Most of us aren't actually hoarding anything. We have largely been conditioned into silence.
For a long stretch, the messaging from official channels was loud and clear: don't use this, it's a risk, it's banned, it's against policy. Understandable, even smart, from a purely legal read of an unfamiliar risk. But loud.
The reversal has been nothing like that. No press release walked it back. No policy update announced the change of heart. People just started using it anyway, and organizations quietly stopped enforcing what they used to forbid. Acceptance arrived begrudgingly and unofficially, while the ban was public and confident.
So now you've got a population of people who taught themselves this skill in secret, sometimes explicitly against their employer's wishes, and the same institutions that once treated it as a fireable offense are surprised that nobody's rushing to share what they learned. That's not greed. That's a completely rational response to the signal you sent them for two straight years.
What governance actually needs to be
If your organization is finally getting serious about AI policy, good. But if that policy is more than six months old, it's already stale. This moves too fast for anything static to hold.
Real governance needs real buy-in from the top, with legal sign-off that says the benefit clearly outweighs the risk, not a policy written to protect the company from a headline.
Here's what I wouldn't do: mandate disclosure of AI use, or build special review processes for anything AI-touched before it goes external. Those sound responsible. They're not. They treat AI as the variable that needs watching, when the actual variable was always culture. AI doesn't create carelessness, it amplifies whatever was already there. A team that owns its output will own AI-assisted output the same way it owns everything else. A team that doesn't will produce sloppy work with or without a model in the loop, and no disclosure checkbox fixes that.
The standard is simpler than a policy document: you own it. Full stop. However it got made.
A good set of policies should be about raising the floor of quality. That means clearer thinking, better review habits, and safer defaults. Or more plainly, a standard that isn't embarrassing.
Raising the floor looks different depending on the layer
The more effective early adopters started by chasing perfect prompts. We categorized them, treasured them, and then watched them become largely irrelevant with the next jump in model capability.
As we learned there was no single prompt to rule them all, we started building better harnesses around the processes that actually survived the rapid release cycles of frontier models.
I built DialogDrive as a personal project, mostly to solve a problem I used to have when I was a small operator selling into enterprise shops. It's a harness for outreach and research work, tuned to produce sharper, more targeted results than a generic prompt ever will on its own. That's raising the floor on execution: given a task, get the best possible result, repeatably. Because while I can copy and paste a prompt over and over again with reliable enough precision, scaling that takes a harness.
Later I built something else, a tool I call Codex Designer. It is just an open-source wrapper around OpenAI's Codex coding tool. Its job is making sure the person behind the prompt has actually thought through the shape of a system before anything gets built. That discipline, the process of forcing myself to answer every uncomfortable question before I'm allowed to write code, is the actual methodology. Codex Designer is just the tool I built to run it. And Codex Designer itself was built using Codex Designer. Very Ouroboros.
Two tools, two layers. One raises the floor on what you build. One raises the floor on how well you thought about it first.
Culture is a huge driver here
If you want people to stop hoarding, compensate the thing you're asking them to hand over. That doesn't have to mean a raise, though it can. It can be a promotion, a new title that reflects what they've actually been doing quietly for two years, or simply loud, public acknowledgment that this person saw something before almost anyone else did.
Compensation doesn't have to cost the company a dime, although it normally lands better when it has a budget line item attached to it.
The skepticism was public. The credit has to be at least as public, or the asymmetry just repeats itself with the roles reversed.
What good actually looks like
None of this requires you to already be somewhere good. Good is available to any organization, starting from wherever it currently stands. It isn't complicated. Decide what you actually value here. Write it down. Say it loudly enough that the people who took the risk early feel like the risk was seen. So much time was spent forcing early adopters into the dark, equal effort must be spent reassuring that it is ok to come out into the light again.
The more organizations do those simple things, the fewer digital dragon hoards there will be.