Last author on LLM-agent evaluation research: how token budgets, planning information, and release control create value
2
papers here
0
reviewed by Evolve
0
first-author
0
corresponding
—
works · OpenAlex
13
h-index · OpenAlex
Bibliometric figures come from OpenAlex (2026-09-27) and cover the author's entire public record, not just the papers listed here. Reproducibility signal comes from Evolve's own reviews.
Research profile
Output—
Not yet scored: per-work venue, publication type, and total career output counts are unavailable for this record. This axis would measure annual publication volume and output consistency across the researcher's full career.
Impact—
Not yet scored. Citation-linked aggregates beyond the recorded h-index of 13 and i10-index of 15 are insufficient to compute this axis, which would measure citation accumulation and influence relative to field norms.
Focus—
Not yet scored. The record's topic mix is heterogeneous — biomedical spectroscopy, chemometrics, and pain research alongside LLM-agent work — and field-distribution data is incomplete; this axis would measure topical concentration and coherence.
Collaboration32/100
Scores 32/100 on Evolve's collaboration axis. The OpenAlex record lists 4 distinct coauthors in the recent three-year window against recent3yWorks of 5 — a small, stable circle rather than a broad network. Yukun Zhang and Kemu Xu recur as frequent coauthors, and both 2026 catalog papers are last-authored team efforts built around this same core pair. Co-authorship intensity sits well below highly networked peers.
Activity52/100
Scores 52/100 on the activity axis. The record's countsByYear span 2007–2026 with a clear peak of 8 works in 2021, followed by thinner years (3 in 2022, 1 each in 2023–2025) and 3 works already in 2026. recent3yWorks stands at 5, indicating moderate but ongoing output currently concentrated in the new LLM-agent preprints.
Reproducibility—
Not yet scored at profile level. Evolve's per-paper repro-readiness ratings (medium for both reviewed 2026 preprints) exist, but the benchmark artifacts and code-availability evidence this axis requires have not yet been compiled for the full record.
Research summary
Yishen Chen is a name-resolved OpenAlex profile whose reviewed work centers on the economics and engineering of LLM-agent systems. As last author on two 2026 arXiv preprints — "The Organization of Inference" and "How Do Agent Harnesses Create Value?" — Chen studies how planning information, token budgets, and release control shape what agent harnesses actually add, using preregistered SWE-bench Verified experiments and sham-controlled tau2-bench evaluations (a reported 7.17 pp success lift over word-matched controls). The bibliometric record shows an h-index of 13 and i10-index of 15 across a career spanning 2007–2026, with peak output of 8 works in 2021, 5 works in the last three years, and 4 distinct coauthors, with Yukun Zhang and Kemu Xu as the recurring collaborators. Notably, the record's dominant OpenAlex topics (biomedical spectroscopy, chemometrics, pain research) differ sharply from the agent papers reviewed here, so this profile may mix identities or capture a recent pivot; either way, the reviewed evidence base is the two 2026 preprints. Evolve rates both as medium repro-readiness, crediting the preregistered design in one while flagging narrow tasks, drifting API backends, and a statistically fragile headline effect in the other.
Collaboration pattern
Collaboration is compact: the record lists 4 distinct coauthors in the recent window (collaboration score 32/100), with Yukun Zhang and Kemu Xu recurring as the stable core team. Both reviewed 2026 preprints are last-authored, non-corresponding team papers built around this same pair. The pattern suggests a small, consistent working group rather than a wide collaboration network, and no cross-institution partners are visible in the provided data.
Research keywords
Highlights
Token budgets frame the value of planning in AI production
Prewritten plans lift agent success by 7.17 percentage points
Timeline
- 2007Earliest publication year in the OpenAlex record (1 work) — the start of a research record that now carries an h-index of 13.
- 2016The record shows 2 works in 2016, the start of a two-year stretch at that cadence.
- 2021Peak output year: 8 works recorded in a single year, the highest annual count in the record.
- 20223 works recorded as annual output tapers from the 2021 peak.
- 2026Two arXiv preprints on LLM-agent systems — both last-authored — enter the Evolve catalog, marking a visible turn toward AI-production research.
Papers in the Evolve catalog
Frequent co-authors
About this profile
This profile is AI-structured by Evolve.Pub from public bibliographic data (OpenAlex) and Evolve's own paper reviews. It is not written or endorsed by the researcher unless the profile is claimed. Radar dimensions are computed from real bibliometrics with fixed formulas — raw figures are shown alongside; missing dimensions mean insufficient data, never zero. Errors? Claim the profile to correct or hide it.
- Bibliometrics are name-resolved from a single OpenAlex record with no ORCID or affiliation available to confirm identity; the record's dominant topics (biomedical spectroscopy, chemometrics, pain research) differ from the reviewed LLM-agent papers, so entity mixing or a recent research pivot cannot be excluded.
- Output and impact dimensions are unscored because per-paper citation counts, venues, and total publication counts were not provided, even though h-index (13) and i10-index (15) appear on the record.
- The Evolve evidence base covers only two 2026 arXiv preprints, both non-corresponding last-author works, so the catalog view reflects the researcher's most recent output rather than the full career.
- The reproducibility axis is unscored at profile level; only per-paper repro-readiness ratings (medium for both reviewed papers) are available.