EGE ERDIL

Behavioral Archetype

THE GROWTH ECONOMIST WHO BUILDS THE ENVIRONMENT — Subject spent his Epoch AI years writing the economics of AI: increasing returns to compute, the arguments for explosive growth, inference costs. He then co-founded Mechanize, a company that sells the training environments and evaluations the models learn from, and says in its essays that the economy’s full automation is coming whether or not anyone chooses it. The forecast and the product point the same way, and the company’s own pages say so. This profile scores the position, not the man.

Essence Indicators

• Co-founder, Mechanize, Inc. Mechanize’s announcement, dated April 17, 2025, lists “Matthew Barnett, Tamay Besiroglu, and Ege Erdil” as the founders and gives no titles. It describes a company “developing virtual work environments, benchmarks, and training data that will enable the full automation of the economy” (Mechanize). The same list of founders appears on its April 24, 2026 funding release (Mechanize). Co-founder Tamay Besiroglu announced the company the same day on X (his post).

• Former researcher, Epoch AI. His author bio on Epoch’s site reads “Ege Erdil is a former researcher at Epoch AI. He has interests in mathematics, statistics, economics and forecasting,” on a post dated March 7, 2025 (Epoch AI). He is not on Epoch’s current team page (Epoch AI). Epoch’s own page gives no start or end date; none is asserted here.

• Epoch authorship. Co-author with Besiroglu of “Explosive growth from AI: A review of the arguments” (September 23, 2023), which concludes “the odds likely stand perhaps slightly better than even for automation of the sort facilitated by advanced AI” (Epoch AI), and of “Train once, deploy many,” whose thesis line is “AI systems exhibit increasing returns to scale when we add more compute for training and inference” (Epoch AI).

• Mechanize’s current product line. “We build environments and evals for frontier coding agents” (Mechanize). Its hiring page lists a “Research Engineer, Alignment” whose goal is “Build evals that test for misaligned model behaviors” (Mechanize). Mechanize reports it “raised $9.1 million at a $500 million post-money valuation” (Mechanize).

• Jaime Sevilla’s line against the venture. Epoch’s CEO Jaime Sevilla wrote in April 2025 that he was “NOT joining Mechanize” and later “I have no personal stake in Mechanize” (his post; his post).

Social Persona / Impression Management

Immediate impression: The quantitative one of the three founders: the economist whose arguments carry the company’s essays and whose conversational register is arithmetic, elasticities and wage curves.

Energy: Thesis-driven and unhurried. On the Dwarkesh podcast, April 17, 2025: “The fundamental thing is that there’s no reason fundamentally why economic growth can’t be much faster than it is today” (Dwarkesh Podcast, 01:24:59).

Impression management strategy: Answer the objection on its own terms. The company’s November 2025 essay opens “Our critics tell us that our work will destroy the world” and replies at length (Mechanize). The essays are bylined jointly, so the position is the firm’s and each author’s.

Forensic Archetype Comparison

Psychometric Assessment

Trait scores observe the documented public role, not private character.

Big Five (OCEAN):

Dark Triad (the score measures the public role, not character):

MBTI: INTP-adjacent, inferred from the documented public role only. A model-first, elasticity-first register; opinions argued from first principles.

Threat Assessment

No prior scoring; this is a first. Calibrated against peers: Tamay Besiroglu (6/6/7, hazard 62, the co-founder who announced the company and is the public face), and Jaime Sevilla (5/7/7, hazard 63, CEO of the organization both left). He takes attack 6 and defense 6 like Besiroglu (the company is the actor; his own disclosures are thinner) and hp 6 where Besiroglu takes 7, because no executive title, board seat or personal channel is documented for him and the record rests on joint bylines. Hazard follows the (atk+def+hp)x10/3 rule: 60.

Alignment Analysis

Stated alignment: Accelerate automation because the growth is worth it. The company’s essay: “Autonomous agents that fully substitute for human labor will inevitably be created,” and on the podcast, as the host quotes the announcement and he defends it: the case is “we’re going to have this enormous increase in economic growth” (Mechanize; Dwarkesh Podcast, 02:12:49). On timelines, April 2025: “Maybe for me, that would be around 2045” (Dwarkesh Podcast, 00:02:05). On safety work: “the scaling is just really important to make progress in alignment, in our view” (Dwarkesh Podcast, 02:15:18).

Observed alignment: The company sells “environments and evals for frontier coding agents,” employs engineers to “Build evals that test for misaligned model behaviors,” and raised $9.1 million at a $500 million post-money valuation from named individual investors (Mechanize; Mechanize; Mechanize). Its announcement lists backers including Nat Friedman, Daniel Gross, Patrick Collison, Dwarkesh Patel, Sholto Douglas and Marcus Abramovitch (Mechanize).

Gap assessment: No gap in candor is documented: the aim is stated and the product follows from it. Two gaps are structural. The first: the founders came from the measurement layer and now sell inputs to the thing measured, while also hiring for misalignment evals, so the same firm is a source and a tester of capability. The second: Mechanize’s earliest copy said “full automation of the economy,” and its current homepage says “environments and evals for frontier coding agents”; the narrower line is the product, the broader line is the essays. The record settles both; it does not settle whether either affects outcomes.

Interest. Each position is in his or the company’s own words. What adoption would pay is read from Mechanize’s pages. None of it is motive and none of it says the argument is wrong.

The convergence is the finding. The case for hastening automation, for not expecting a coordinated stop, and for scaling as the path to safety is made by the firm that sells the training environments, and the same firm lists alignment evals as a product line.

Alliance Map

Overlap: the ties run at the same time on one axis, Corporate. He sits between a lab-funded measurement nonprofit he has left and a supplier whose likely buyers are the labs the nonprofit measures, with named investors from the lab and operator class. Nothing is concealed; the company’s pages state it. That is the finding, and it holds without any claim about influence.

Convergent Drive Classification

Self-preservation: Moved from a nonprofit to a funded company, then raised a round at a stated $500 million valuation (Mechanize). Goal preservation: One stated aim, held across the 2025 announcement and the 2026 essays: the full automation of the economy (Mechanize). Resource acquisition: Backers from the top of the technology investor class, listed by name (Mechanize). Self-improvement: From forecasting automation’s economics to building the training ground for it (Mechanize: “our immediate goal is to remove this bottleneck and accelerate progress toward a fully automated economy”).

Subject is not an AI system. The drives appear anyway — in the economist who wrote the growth curve and then opened the shop that sells the ground it grows in.

Sources: Announcing Mechanize, Inc. — Mechanize; Mechanize raises $9.1M — Mechanize; Mechanize — home; Apply — Mechanize; The future of AI is already written — Mechanize; Life after work — Mechanize; Unfalsifiable stories of doom — Mechanize; How to fully automate software engineering — Mechanize; AGI is still 30 years away — Dwarkesh Podcast; Train once, deploy many — Epoch AI; Explosive growth from AI — Epoch AI; Besiroglu announcing Mechanize — X; Sevilla — X.