The Vampiric Effect

The vampiric effect: you produce 100x output in three hours, and then you're expected to work eight hours. So you produce 300x. And then they recalibrate again.

Here's the answer up front: AI productivity gains don't automatically go to you. They go to the expectation machine. Unless you structure the exchange before you demonstrate what's possible, the gain evaporates — and it takes something with it.

Steve Yegge described it in a Pragmatic Engineer conversation in May 2026. I'd heard versions of this idea before, but never with that exact word. Vampiric. It's right. You generate, they absorb. You generate more, they absorb more. The cycle doesn't break unless you break it.

Why this matters

When an engineer 3x'd their output in 1995, there was a lag. The feedback loop between "I can do more" and "now you're expected to" was slow enough that people had time to negotiate. Raises, promotions, titles — the system had time to process.

AI is different. The speed of the gain is faster than the speed of renegotiation. By the time you sit down to ask for more, you've already produced twice as much and the baseline has quietly shifted under you.

The engineers working with agentic coding tools right now are hitting 100x output windows. Three hours of real focused work. It doesn't feel sustainable because it isn't — not at the cognitive load it takes to orchestrate AI systems well. It's not running a faster car. It's running a fleet while also driving the lead car.

What actually happens

The gain is real. I'm not arguing that. What I'm saying is that productivity is only worth what you capture from it.

If you work somewhere that absorbs your output into their margin without adjustment — and most workplaces will, because that's the default — you've just automated your own leverage away. You did the hard work of figuring out how to 10x your output, and the reward was a higher bar.

The people who thrive in this next cycle will be the ones who negotiated before they demonstrated. Who said: if I can deliver X using these tools, I want Y — before they showed anyone what X looks like. Once it's visible, the baseline has moved. You can't un-demonstrate capability.

This isn't cynical. It's structural. Price it before you ship it.

From my own bench

I run agentic workflows after hours while holding down a day job. I've had weekends where I shipped more than I used to ship in a month.

The temptation is to just keep going. Let the ceiling rise. Produce more because you can. I've fallen into it.

But more output without more leverage is a treadmill. And agentic treadmills run faster than human ones.

What I've learned the hard way: show the result once, price it, don't let "what you can do" become the free floor for "what they expect." The first demonstration sets the anchor. Choose your anchor carefully.

There's also the energy piece, which people skip. Orchestrating AI systems is real cognitive work. Running agents, reviewing output, catching failures, verifying decisions — this has fatigue attached to it. If you sprint on AI tooling without building recovery into the structure, you'll hit a wall. The wall doesn't care about your throughput metrics.

Try it today

Step	What you do	Why it pays off
1. Measure the delta	Track your AI-augmented output for two weeks in concrete terms — not vibes, actual deliverables	You can't negotiate what you haven't measured
2. Find one anchor	Identify one deliverable that took 2 hours that used to take 2 days	That delta is the opening number for any compensation conversation
3. Name the price first	Have the conversation before you demonstrate publicly	Once everyone sees what's possible, the baseline has already moved

Where people get burned

• Producing more and not capturing the gain. You get tired, they get value, the exchange stays lopsided. Fix: Treat your productivity improvement like a product feature — price before you ship.
• Burning out before you can negotiate. Orchestrating AI systems drains cognitive energy. If you sprint without recovery built in, you'll wall out. Fix: Cap your agentic work hours the same way you'd cap intense exercise.
• Showing everything you can do in one meeting. Now the bar is permanently there. Fix: Reveal capability incrementally and attach a value exchange to each reveal.
• Assuming your employer will notice and reward it unprompted. They won't. Fix: Name the gain explicitly. "I delivered X in the time it used to take for Y. I'd like to talk about what that's worth."

Tools, and a question worth sitting with

• A thing to try: Run a two-week log — deliverable, estimated old time, actual new time. Don't share it with anyone yet. Use it to find your strongest anchor before you have the conversation.
• A watch: Steve Yegge's conversation with the Pragmatic Engineer is worth your time if you're in software. The vampiric effect framing starts around halfway through. He's not alarmist about it — which makes it more useful.
• A question to actually sit with: If you doubled your output this quarter, did you capture any of that value — or did you just set a new baseline?

The bottom line

AI makes you faster. That part is real. What isn't automatic is the value exchange. The productivity gain is yours to keep only if you structure it that way before you demonstrate it.

The teams that will survive this cycle aren't the ones who produce the most. They're the ones who figured out how to capture what they produce.

— Dru Edwards