Nicholas Arcolano, Ph.D.

Chief Scientist / Jellyfish

The biggest barrier to AI adoption isn't the technology.

It's measurement, organizational readiness, and knowing what's actually working. I'm a researcher and advisor focused on AI's real-world impact on software engineering.

In my role as Chief Scientist at Jellyfish, my team and I study AI adoption using the largest engineering dataset in the industry: 1,300+ companies and more than 100 million pull requests. We turn what we find into guidance for leaders putting AI into practice.

in/arcolano @arcolano [email protected]

selected

talkAI Engineer New York · 2026-10 It's Not the Model, It's You podcastAI Native Dev · 2026-06 Why Developers Hit a Wall at 4 AI Agents postTessl · 2026-10 There Aren’t Enough Chaperones for All Your Coding Agents pressTechCrunch · 2026-06 The Token Bill Comes Due: Inside the Industry Scramble to Manage AI's Runaway Costs

01 / Key Findings

Highlights from my research with Jellyfish, drawn from engineering workflow data across 1,300+ companies, roughly 300,000 engineers, and more than 100 million pull requests.

  • Agents are writing more code than teams can trust. At the median company, autonomous coding agents authored 22% of merged PRs (53% at the 90th percentile), but only about 0.6% of PRs were merged by agents without human review. Agent-authored PRs are abandoned at twice the rate of human ones (40% vs. 20%). Tessl, 2026-10
  • More PRs aren’t the same as more product. Over 2026, pull requests roughly doubled while product outputs like features shipped rose only about 40–50%. Tessl, 2026-10
  • The model barely matters; the engineer matters most. The AI tool itself (model and harness) explains less than 1% of the difference in what engineers ship. AI usage volume explains about 9%, company and team about 12%, and the individual engineer more than 75%. The Daily Context, 2026-10
  • Tokenmaxxing hits diminishing returns. The top 10% of Claude Code users consumed about 10x the tokens of the median developer but produced only about 2x the output, and cost per merged PR rose from $0.28 at the lowest adoption tier to $89.32 at the highest. TechRadar Pro, 2026-07
  • AI-assisted PRs are bigger. Going from 0% to 100% AI adoption corresponds to about 18% more lines added per pull request. Jellyfish, 2025-09

02 / Writing & Research

I write data-driven research on AI and software engineering. Recent pieces I've authored:

show all writing (14 more)

subscribe → jellyfishresearch.substack.com

03 / Speaking

Upcoming appearances, plus recent talks and panel discussions on AI and software engineering. Several are available to watch online.

upcoming

talks

panels

  • AI Engineering Trends: Research PanelJellyfish AI Impact Week · 2026-10
  • AI in SDLC: Executive Panel & Hands-on WorkshopAWS · Boston · 2026-07
  • Leaders on AI's Impact on the Development Life CycleAWS · Los Angeles · 2026-06
  • From Copilots to Agents: The New Era of Software DeliveryJellyfish & Pillar VC · Boston · 2026-03
  • AI in Action: Modernizing the SDLCAWS · Palo Alto · 2026-02
show all panels (2 more)

04 / Podcasts

  • Why Developers Hit a Wall at 4 AI AgentsAI Native Dev (Tessl) · 2026-06youtube · spotify · apple
  • Behind the Scenes of AI Coding AssistantsParadigm Shock · 2026-01spotify · apple

05 / In the Press

Selected press coverage of my work on AI and software engineering.

show all press (5 more)

06 / Get in Touch

I post regularly on LinkedIn and X about AI and software engineering. I also advise early-stage founders as a founding member of a16z speedrun's Insiders program. For everything else, email is the best way to reach me.

linkedin.com/in/arcolano x.com/arcolano [email protected]

for organizers & press

For speaking invitations, interviews, and press requests, email [email protected]. A high-resolution headshot is available to download.

  • Topics: measuring AI’s real impact on engineering productivity · autonomous coding agents, trust, and code review · token economics and the cost of AI · why people and practices matter more than model choice · leading organizations through AI adoption
  • Short bio: Nicholas Arcolano, Ph.D., is Chief Scientist at Jellyfish, where he leads research on how AI is changing the way software gets built, drawing on workflow data from 1,300+ companies and more than 100 million pull requests.
  • Long bio: Nicholas Arcolano, Ph.D., is Chief Scientist at Jellyfish, where he leads Jellyfish Research. His team studies AI adoption using the largest engineering dataset in the industry, covering 1,300+ companies, roughly 300,000 engineers, and more than 100 million pull requests, and turns its findings into guidance for engineering leaders. His work has been covered by TechCrunch, Business Insider, LeadDev, and The New Stack, and he speaks regularly at events including AI Engineer and PlatformCon. Before Jellyfish, he led data science at TrueMotion and Runkeeper, and spent a decade at MIT Lincoln Laboratory researching statistical and machine learning methods. He holds a Ph.D. in applied mathematics from Harvard University and is a founding member of a16z speedrun’s Insiders program.