Suggestions
Rob Leclerc, PhD
Deeptech and deep science VC
Rob Leclerc, PhD is a founding partner of AgFunder, a leading foodtech and agtech-focused venture capital firm.1 He joined Oper8r LLC as a Cohort 2 Member in April 2021, which is a curated community building the next generation of institutional VC firms.2
Leclerc has an extensive background in food and agriculture technology, with over 15 years of experience as a scientist and engineer. He holds a PhD from Yale University in Computational and Theoretical Evolutionary Biology.12
As a prominent figure in the agtech industry, Leclerc:
- Contributes to Forbes as a writer2
- Has been published in TechCrunch, Harvard Business Review, and Nature journals1
- Speaks at global conferences on food, agriculture, technology, and fintech1
- Has been recognized by Forbes Magazine and Progressive Farming on their 'Up-And-Comers' lists1
Leclerc is also an active investor in various food and agriculture technology startups, including Bite Ninja, Bear Flag Robotics, IntegriCulture, and Atomo Coffee.2 His expertise spans areas such as artificial intelligence, virtual robotics, and computational biology.23
Highlights
x has become a private journal with public permission.
i had codex do a gap analysis on my openclaw-based system and codex. interesting analysis and shows some serious gaps in openclaw/PI agency. good argument for calling codex cli directly in openclaw:
-
Long multi-step objectives: “go do this, then verify, then report” may still stop earlier than Codex.
-
Follow-up continuation: “yes, do it” can still be fragile unless the active objective is preserved and resumed.
-
Tool failure recovery: Codex tends to inspect the failure and try another route; your PI may convert failure into an answer.
-
Browser/web tasks: PI may have browser access, but the loop around “click/type/inspect/submit/block” is not yet as native as Codex’s computer/workspace posture. Repo-edit objectives: Codex has first-class edit/test/commit muscle. Your PI can call into that, but it is still mediated.
-
Compaction/resume: Codex has stronger built-in continuity. OpenClaw has to model that explicitly.
-
Truthful terminality: We added a gate, but Codex’s whole harness is naturally organized around “keep working until done or blocked.”

