Compensation and equity for hybrid human-AI teams

"Do the engineers using Cursor heavily get paid more, or is that a slippery slope?"

That's the question I keep getting from founders 6 weeks out from a comp cycle. They have engineers shipping 1.5x to 2x output with AI tools and engineers shipping less without them. They have no model that handles this. Their People Ops lead is asking for guidance and the comp consultants are pitching theoretical frameworks that don't survive contact with a real review.

This is the operational answer. Five questions. Real examples. What actually changes about comp, leveling, and equity when an agent does 30 to 40% of a role.

TL;DR

The 5 questions to ask before the comp cycle

Walk into your next comp meeting with these 5 questions on the agenda. They surface the AI conversation without forcing a premature framework.

1. Did our productivity definition change in the last 12 months? If the team is shipping 30% more code, closing 30% more tickets, or generating 30% more pipeline, the definition of "meets expectations" at each level is now higher. You can pay the same dollars and require more output. That's not a comp cut. That's a leveling recalibration.

2. Are we measuring output or outcomes? Output-based comp (lines of code, tickets closed, meetings booked) breaks immediately in an agentic environment because the agent does the output. Outcomes-based comp (architecture quality, customer retention, deal close rates) still works. If your performance system is output-based today, you have a structural problem to fix before you talk about AI specifically.

3. Are the people who use AI well promoting faster than the people who don't? If yes, that's healthy if it's about judgment and outcomes. If it's just about tool fluency, you have a leveling problem. Senior engineering should not be a Cursor proficiency exam.

4. Are we paying the same bands as we were 12 months ago, or are local markets adjusting? Comp bands are set by local market and competitive dynamics, not internal philosophy. If everyone in your geography is paying staff engineers $40K more this year, you adjust. The AI question is what you're getting for that adjusted band, not whether to make the adjustment.

5. What's our equity philosophy? Equity is the riskiest place to "AI-adjust." It's also the place most teams are tempted. Resist. We'll get into this in detail below.

Why output-based comp breaks

Three engineers on a 6-person team. Same level. Same band.

Engineer A uses Cursor and an AI code reviewer heavily. Ships 22 PRs this quarter. 18 of them merged in under a week.

Engineer B uses AI tools occasionally. Ships 12 PRs. All of them complex architecture work.

Engineer C is skeptical of AI tools. Ships 9 PRs. Mentors 3 juniors. Runs the technical interview loop. Is the person every other IC goes to with a design question.

If you tie comp to PR count, you pay Engineer A the most. After 6 months, Engineer C leaves because they got a smaller bonus than the engineer who knows how to write a Cursor prompt. The team loses its senior IC bench. The system collapses.

If you tie comp to actual outcomes (architectural soundness, mentorship, design ownership), the order flips. Engineer C is paid most, Engineer B is solid, Engineer A is doing fine and probably needs a leveling conversation about what mid-to-senior looks like in their next promo cycle.

The principle: AI changes which work is easy. It does not change which work matters. Comp tied to what matters survives. Comp tied to what's easy collapses.

What "meets expectations" looks like one year in

The honest update to performance leveling is structural. Here's what we've seen work.

Junior engineer (L3 / E3). Used to mean "writes correct code, passes review." Now means "writes correct code with AI assist, understands when the AI is wrong, asks the right questions in review." Same band. Higher implicit standard.

Mid-level (L4 / E4). Used to mean "ships features end-to-end." Now means "ships features end-to-end and decides intelligently which sub-tasks to delegate to agents." The judgment about what to delegate is the new mid-level skill.

Senior (L5 / E5). Used to mean "owns systems, mentors juniors, reviews 8 to 12 PRs a week." Now means "owns systems, mentors juniors on AI delegation, sets architecture that agents can extend safely, reviews the 30% of PRs that AI cannot adequately review." Same band. Different job.

Staff (L6 / E6). Used to mean "owns cross-team initiatives, sets technical direction." Now means "owns cross-team initiatives, sets the team's AI tooling and orchestrator strategy, builds the verification layer that catches what agents miss." Same band. Bigger scope.

Notice what didn't change: comp bands. Staff still makes staff money. The work each level is responsible for shifted. Internal promotion velocity carries the differentiation across ICs at the same level.

This is the path most companies should take. It's also the boring path that doesn't make headlines. The headlines belong to companies that paid one engineer $400K more than peers because they "5x'd their output with AI." Those companies are going to have a turnover problem in 9 months.

The equity conversation

Equity is the place where most founders are tempted to "AI-adjust" and the place where doing so causes the most damage.

Equity is not paying you for the output you produce in a given quarter. Equity is paying you for showing up, owning outcomes, and helping the company grow over a 4-year horizon. It rewards ownership, judgment, leadership, retention, and risk tolerance. None of those scale with AI tool fluency.

Three traps to avoid:

Trap 1: Refresh grants tied to AI-augmented output. Don't. You'll incentivize the wrong work. Engineers will optimize for AI-shippable PRs over architecture work.

Trap 2: Bigger initial grants for "AI-native" hires. Tempting because they ship faster early. Bad because the AI-native skill curve flattens fast. After 6 months, every IC on your team is AI-native or close to it. Now you have a permanent grant differential for a skill that's become table stakes.

Trap 3: Smaller grants because "the agent does some of the work." Equally bad. The agent doesn't take equity. The human still bears the risk of an early-stage company. You owe the equity for that risk and that long-horizon ownership, not for the quarterly output.

Keep your equity philosophy intact. Tie it to level, tenure, and the ownership the role requires. Let the comp band hold the differentiation by role, and let promotion velocity hold the differentiation between ICs.

The performance review needs two new conversations

The traditional review has the manager and the IC discuss what they shipped, what they learned, what's next. That structure still works. Add two specific conversations to it.

Conversation 1: "What did you delegate to agents this quarter that actually worked?" This surfaces the judgment skill. ICs who can articulate where they used agents well and where they pulled back are the ones leveling up correctly. ICs who can't articulate this are either using agents indiscriminately or not using them at all. Both are leveling concerns.

Conversation 2: "What did you keep that an agent could have handled?" This is the more important conversation. ICs who keep work they could have delegated are bottlenecking themselves and the team. ICs who can name specific work they should have delegated and didn't are showing the right meta-cognition. This is the senior IC promotion signal.

The two conversations together replace what used to be a single "how productive were you" conversation. Productivity is still part of the review. The shape of how you got there is now a separate axis.

Two real comp-band examples

The two patterns we've seen work in companies running this conversation for 12+ months.

Pattern A: Hold bands steady, raise the floor at each level. Same dollars. Higher expected output and quality at each level. Promotion velocity carries the differentiation. Used by most successful AI-fluent engineering teams we know of. Boring. Effective.

Pattern B: Modest upward band shift across the org. Raise all bands by 8 to 12% to reflect the higher market for top talent, raise expectations correspondingly. Used by companies in hot local markets where comp competition is fierce. More expensive. Necessary when you're competing for the same 200 senior ICs as Anthropic and OpenAI.

Pattern we have not seen work: variable comp tied to AI-augmented output. Tried by 3 companies we know of in 2025. All 3 walked it back inside 18 months. The behavior incentives went sideways. The team culture suffered. They reverted to outcome-based reviews with stable bands.

For the broader pattern of how comp fits the agentic-AI-first org structure, see the cornerstone post.

What changes about hiring

Three honest updates to your hiring loop.

Tool fluency in the screening. Add 10 minutes to the technical interview where the candidate uses Cursor or Claude live. Watch how they delegate, what they verify, what they push back on. This is a competency now. Don't make it the whole interview, but don't skip it.

Leveling at offer. Some candidates who would have been mid-level 18 months ago are now senior-bench because they have the AI-delegation judgment. Some senior candidates from large companies are mid-level because they have never had to make AI-delegation calls. Calibrate.

Compensation for AI-augmented roles. Pay the role, not the augmentation. A senior engineer who ships 2x with AI assist still gets senior engineer comp. The 2x is built into the bar for the level, not into the offer.

Common failure modes

The mistakes we see often enough to call them patterns.

Two-class engineering team. AI-fluent ICs promote faster, AI-resistant ICs feel undervalued, the AI-resistant ICs quit. Solution: make the leveling conversation explicit and early. "Here's what senior looks like in 2026. Here's what we need from you to get there. Tool fluency is one input, judgment is the bigger one."

Output dashboards everywhere. Once you have AI-augmented output, the temptation is to track everything. Track 3 things. PR throughput, bug-escape rate, architecture-debt change. The rest is noise that tilts incentives the wrong way.

No comp conversation at all. Most teams just keep operating as if nothing changed and hope the comp question doesn't surface. It surfaces. It usually surfaces in an exit interview from a senior IC who felt invisible. Have the conversation proactively.

Cutting headcount to "fund AI tooling." $200/month per IC for AI tools is rounding error compared to the salary cost. Don't trade headcount for tooling budget. Trade headcount for output expectations.

What to do in the next 14 days

The realistic sequencing.

Day 1. Map your current engineering team against the new leveling expectations above. Where is each IC? Where do you want them to be in 12 months?

Day 2. Audit your performance review template. Does it have the 2 new conversations? Does it still tie comp to outcomes, not output? Adjust.

Day 3. Talk to your People Ops lead. Walk through the 5 questions. Pick the answers your company is going to commit to publicly within the team.

Days 4 to 10. Brief the management chain. Make sure every manager has the same answer to "why aren't we just paying the AI-heavy engineers more." It will come up.

Days 11 to 14. Schedule the team conversation. Communicate the leveling shift. Be explicit about what's changing about expectations and what's staying the same about comp. Take questions.

Month 2. Run a comp cycle with the new framework. Measure how it lands. Iterate.

That's the path. Not "AI-adjusted comp consultancy engagement." Not "new equity grants tied to tool usage." Operationalize the leveling shift, hold the comp bands, let promotion velocity carry the differentiation.

Next up

Final post in this batch: vendor selection for agentic AI. Build vs buy vs orchestrate. The decision framework, the cost comparison, and how to avoid the 18-month enterprise procurement trap. Then we'll get back to the cornerstone posts and the lived-experience deep dives.

If you want the leveling and comp framework built for your team in the next two weeks, see the blueprint catalog or email christine@operatoriq.io. Email only, no calls.

Cheers, Christine