How this research is made
Every brief on Agentic Engineers is produced the same way: deep research conducted by AI research agents, then reviewed for accuracy by a human editor. This page explains that process, the source-credibility standard the work follows, how we treat vendor claims, how we verify and date the material, and how to report a correction. The goal is simple — readers should be able to see where every claim comes from and how much weight to put on it.
AI research, human review
The briefs synthesize deep research carried out by AI research agents. That research is then reviewed by Trish Uhl, who is the editorial reviewer and the named author behind the work published here. Figures and claims are reproduced as reported from the underlying research and read as directional signals, not verified universal benchmarks. Where the analysis is our interpretation rather than a measured fact, we say so — with framing such as “the research argues,” “reported,” or “available cases suggest.”
The source-credibility standard
Every claim is weighed by the strength of its source before it is published. We work in three tiers:
- Primary / authoritative — peer-reviewed or preprint papers that resolve, official standards (NIST, OWASP), first-party vendor documentation, named primary posts, and government data.
- Reputable secondary — established analysts and outlets, and vendor case studies (treated as company-reported, not audited).
- Weak / needs scrutiny — personal blogs, marketing and recruiter pages, social posts, AI-generated pages, single-citation chains, and undated or unverifiable material.
Weaker sources are flagged in place rather than presented as settled fact, the stronger source is preferred when sources conflict, and a striking figure that rests on thin provenance is hedged or attributed where it appears. We do not launder weak claims into confident prose.
How we handle vendor and company claims
Adoption percentages, efficiency gains, and case-study outcomes reported by a company about its own product are presented as company-reported claims, not independently audited results. Where a vendor figure describes a single named engagement, we scope it to that engagement rather than presenting it as a market-wide benchmark.
Verification and fact-checking
Before a citation is published, we trace its provenance and prefer a primary source over a summary of one. Cited preprints are checked to confirm the identifier actually resolves to the paper attributed to it. Statistics framed as study results are scoped to the study's sample rather than generalized, and we separate what a primary source actually said from third-party commentary about it.
Embedded source material
Some briefs include the complete underlying research document inline, behind a “View full source material” toggle, so the analysis can be audited against its source. That appendix is reproduced verbatim. When the reproduced text overstates what its own primary source supports, we do not silently rewrite it — we add a visible editorial note next to the toggle explaining the discrepancy.
Updates, dating, and freshness
Each page carries a published date and a last-reviewed date in its structured data, and a visible “Last reviewed” date in the footer. When a brief is materially revised — for example, to correct an attribution or rescope a figure — the review date is advanced and the change is recorded below.
Revision history
June 2026 — initial publication. First public release of the briefs, the source hub, and this methodology.
June 2026 — source-integrity pass. Following an external source-integrity review, we: corrected the Andrej Karpathy brief to match his actual February 2026 post (he distinguished casual “vibe coding” from professional “agentic engineering” rather than retiring the term, and the failure-framed figures that traced only to social-media commentary were removed); scoped the Deloitte 25–30% procurement figure to its single client example; relabeled the Stanford study as 51 success-selected deployments and added selection-bias caveats; and reframed the GitHub coding-agent adoption figure as a sample estimate rather than an all-of-GitHub finding.
Corrections
Spotted a factual, citation, or accessibility issue? Please report it — corrections are welcome and the review date is advanced when a fix is made. Reach the editor via Trish Uhl on LinkedIn.