Larridin: which engineers convert $213/wk in AI tokens to output
Most B2B companies now have a real AI line item: seats for ChatGPT and Claude, token bills from coding agents, and a growing spend on agents that run without anyone watching. Very few can tell you what that money is producing. Larridin is built to answer exactly that question.
It connects AI usage and spend to the work of people and agents across a business, covering adoption and fluency, engineering performance, and opportunities to automate repeated work. It’s backed by a16z, run by a founder on his sixth company, and it just published benchmark data on what AI coding actually costs per engineer and where the returns stop.
The Benchmark Data: A 10x Spend Spread
Larridin published its first benchmark in August 2026, built from production billing and engineering telemetry. The sample covers engineers who, in the four complete weeks ending August 2, 2026, both merged code and drew billed AI-coding spend.
- The median engineer draws $213 a week in billed AI-coding spend, roughly $920 a month.
- The 90th percentile hits $911 a week, a spread of more than 10x.
- These figures are a floor. Engineers on flat-fee plans consume tokens that never appear on a metered bill, so true consumption runs higher.
Annualize the 90th percentile and you’re near $47,000 a year per engineer in tokens alone. For a 100-person engineering org, the gap between your p25 and p90 engineers is a seven-figure budget question most CFOs can’t see today, because spend is scattered across provider invoices, corporate cards, and personal subscriptions.
Same Tools, Same Prices, 2x the Output
Larridin split engineers into cohorts by how much of their shipped output was AI-attributed. Both AI-native cohorts came from the same company, with the same tools, the same prices, and the same starting spend of about $170 a week. The gap between their curves reflects skill, not budget.
- Deeply AI-native (79% AI-attributed): reached 11.8x output at around $1,300 a week and hadn’t hit a ceiling. At equal spend, this group shipped roughly 2x what partial adopters shipped.
- Partially AI-native (37% AI-attributed): got real returns, but marginal payoff dropped by about half once spend passed roughly $600 a week.
- Low-AI (15% or less AI-attributed): topped out around 1.9x and stayed flat across a 20x spend range, from $21 to $421 a week. Extra dollars bought activity, not additional output.
The practical read for 2027 budget planning: more spend converts to more output only where fluency already exists. Larridin’s guidance is to track your own team’s ROI curve and set review triggers where it levels off, rather than copying another company’s cap.
One important caveat Larridin flags itself: these relationships are associational. High-output engineers may spend more because they ship more, not the other way around. Output is not measured in lines of code. Each merged PR is scored by model-assessed complexity across five levels, discounted for low-quality output and missing tests, and scaled by code churn.
️ What the Platform Does
Larridin started as a discovery tool that ran automated inventories of AI tools in use, including shadow AI employees access outside IT’s view. It has grown into four products:
Spend Intelligence
Token usage, seat licenses, and cloud model costs in one view, tracing every dollar to a team, a tool, or an agent before the next budget review. Built for CFOs and finance teams. Agent spend is the fastest-growing and least-understood line in most AI budgets. It doesn’t map to a seat or a person, which is exactly why it needs its own attribution layer.
AI Impact
Connects what each team spends on AI to the hours it returns, and compares adoption, fluency, and cost per AI hour to show where to invest, where to train, and what to scale. Larridin is deliberate about the math: AI capacity estimates the human-equivalent work AI contributes. It is not automatically time saved, cash returned, or a reduction in headcount. Plenty of vendors will hand you an “hours saved” number and let you present it to your board as savings. Larridin explicitly does not do that.
Developer Intelligence
Connects engineering output, code quality, and delivery to AI spend, and shows where coding agents help and where teams need support. The newest piece is Larridin Router. It scores each coding request and serves a lower-cost model when the task allows, while leaving everything else on the model the developer asked for. The part that separates it from a plain model router: it ties routed sessions to output, quality, defect rate, and cost per task, so you can see whether the discount held up in code review. A developer who needs a specific model for a particular job can pin it, and that session gets reported as pinned rather than counted against the router.
Workflow Intelligence
Maps repeated work from observed activity and moves the best candidates from identified to automated, measured against a captured baseline.
Who’s Using It
Gainsight used Larridin to understand AI tool adoption before making its first enterprise LLM purchase. The lesson is simple: without knowing what your people use today, you don’t know what to buy next. Companies sign a big enterprise license based on a pilot, then find half the org is still on personal accounts of a different tool.
Other named customers include Vertiv, Klaviyo, SurveyMonkey, EcoVadis, TigerConnect, and Sundt. The platform is SOC 2 Type II, GDPR, and HIPAA compliant.
The Founder Background
Russ Fradin has been founding and exiting companies for 30 years, starting with Flycast Communications in 1996. He previously ran Dynamic Signal and Adify, and was at comScore. President Jim Larrison came from Dynamic Signal, Firstup, and comScore. CTO Ameya Kanitkar came from LinkedIn, Coinbase, and Groupon.
The Dynamic Signal story is worth noting. After 18 months and $5-6M in ARR, Fradin concluded customers were happy and paying but the business wasn’t sticky, and walked away from it. The rebuilt company became a $50M ARR employee communications business. A founder who has already shut down millions in ARR because it wasn’t recurring is the right person to build a measurement product, where the entire value is being the system people check every week.
Larridin raised $17M in seed funding led by a16z, with Alex Rampell on the board, alongside Bloomberg Beta, Gradient, Haystack, Homebrew, and Refract. They started building in early 2024 and began selling in August 2024.
The Catch: Pricing, Privacy, and Causation
Three things to know before you book the demo.
- Enterprise-priced. Pricing isn’t public. A competitor’s comparison piece puts enterprise pricing starting around $50,000 a year. Treat that number with caution given the source, but plan for an enterprise sales motion.
- Adoption tracking touches employee monitoring. Usage tracking runs through browser plugins and desktop agents. Larridin supports role-based access and enterprise authentication, and lets you choose the scope of measurement. Decide what you’re measuring, and tell your team before you roll it out. Engineers who feel watched will route around the tooling.
- Correlation isn’t proof. Larridin says this about its own benchmark. Use it to find where spend has stopped converting, then test the fix.
Who Should Actually Look at This
Larridin fits a specific profile. If you’re a 20-person startup where every engineer is deeply AI-native and the founder reads the Anthropic invoice personally, you don’t need it yet. Once you’re past a few hundred people, or once agent spend passes seat spend, you probably do.
- CFOs with an AI line item that doubled this year and no way to attribute it by team, tool, or agent.
- CTOs and VPs of Engineering deciding whether to raise, cap, or reallocate coding-agent budgets for 2027. The benchmark data alone is worth reading before that meeting.
- Heads of AI who need to show the board which departments turned licenses into working capability and which are still paying for access nobody uses.
- Companies running agents in production. Agent spend doesn’t sit on anyone’s seat, and it’s the category most in need of attribution.
Start with the Larridin benchmark on AI coding spend before anything else. Then pick one decision you have to make this quarter, a renewal or a coding-agent rollout, and measure that first.
I’m founder of BizStack at Cagri Sarigoz LLC and a passionate advocate for entrepreneurs.
With over 14 years in tech, marketing, and AI, including my role as Head of SEO at CitizenShipper and co-founder of TaleBot at Intale AI, I’m dedicated to sharing genuine, useful product insights and tips.
At BizStack, I aim to cut through the digital noise to provide clear, actionable advice.
And more than all else, I’m a father to a (always) little girl and a husband.
Contact me at [email protected] for assistance.
