two researchers, same three questions: the agent vs weebo

Visual version with the merged lists and animations: /reports/2026-09-04-two-researchers.html. Raw runs and merged findings are listed at the bottom.

On September 4, 2026, two researchers took the same three briefs, independently and in parallel: the agent (Claude, working from web search) and Weebo (a Grok bot team working from X and primary sources). The briefs: the largest 2026 funding rounds by accelerator alumni, the top bootstrapped SaaS launched in 2025–26 by stated revenue, and the top indie hacker products of 2026 by earnings. This is how the two runs compared, and how they are used together from now on.

Headline numbers

Brief 2, fundraisings. The agent checked 45 candidates, Weebo 42. The two top 10s overlapped on 5 rounds. The merged top 10 has 3 rows only the agent found and 2 only Weebo found. 9 disagreements logged, none on round size. All 10 merged rows have two independent sources.

Brief 3, bootstrapped SaaS. The agent checked 47 candidates, Weebo 32. The two top 10s overlapped on 0 products (2 if the agent's ranks 11–12 count). The merged top 10 has 6 rows only the agent found and 2 only Weebo found. 10 disagreements logged. Only 4 of 10 merged rows have two independent sources.

Brief 4, indie products. The agent checked 43 candidates, Weebo 28. The two top 10s overlapped on 2 products. The merged top 10 has 4 rows only the agent found and 3 only Weebo found. 11 disagreements logged. 6 of 10 merged rows have two independent sources.

Candidate counts are each run's own. In every brief the merged list differs materially from both inputs; in brief 3 the two top 10s are disjoint.

Sources each run used

The agent. For fundraisings: an accelerator cohort file as the seed, TechCrunch's mid-year unicorn roundup and Y Combinator tag, AlleyWatch's monthly "largest US rounds" lists read line by line, Business Wire and PR Newswire, the YC company directory (a 404 was treated as "not YC"), and Techstars' own profiles for non-YC verification. For bootstrapped SaaS: the TrustMRR leaderboard and startup pages (Stripe-connected MRR, founded month, last-sync date), Indie Hackers interviews, Starter Story, Inc., Lenny's Newsletter, Hampton, Wix investor releases; listicles for discovery only. For indie products: Starter Story's 2026 interviews and podcast summaries, Indie Hackers, TrustMRR, games press (NotebookCheck, GamesRadar, IndieGame.com), Steam news posts, raijin.gg, and App Store data via mwm.ai. Common limits: X posts could not be fetched directly (status URLs were taken from search-result titles), and the search budget ran out in all three runs before secondary checks were done; Hacker News, Reddit and Product Hunt sweeps yielded nothing.

Weebo. For fundraisings: company newsrooms and press releases first, then TechCrunch, CNBC, GeekWire, Inc., Tech.eu and Tubefilter, then the YC directory or YC's LinkedIn; it documented a three-hop chain per row and kept an exclusion table for large non-accelerator rounds (Ramp, Modal, Cognition, CuspAI). For bootstrapped SaaS: founder LinkedIn posts, ARRFounder milestone pages (with an explicit "aggregator, confirm against primary" rule), Indie Hackers, independent case studies (Tinybird, MRR Story, Superframeworks, Small Start, Startup Founder Stories), TrustMRR and Vibecrowd spot checks. For indie products: MRR Story, Rork, Setapp, Superframeworks, Small Start, Startup Founder Stories, founder LinkedIn, TrustMRR and RevenueCat mentions. Common limits: fewer candidates per brief, no games, wrappers or extensions covered, one source cited without a URL (the Business Wire correction in brief 2), and a design choice in brief 4 to keep SaaS out of the ranked list.

Where each was stronger

Eligibility discipline. The agent applied the rules more strictly and more consistently: it separated "went through the program" from "backed by" (Baseten, Lovable), counted equity only (Astranis, and would have cut Regent to $120m), rejected relaunches and rebrands on the launch-year test (Kleo, SuperX, Zigpoll, Postiz in brief 3), and verified Preply's Techstars class against Techstars' own profile and 2015 press. Weebo was looser: it ranked Regent at its $240m headline with half of it debt, ranked Zigpoll (a 2018–19 product) as a 2026 breakout, accepted Kleo's rebuild as a 2025 launch, and listed Railway as Techstars on "alumni lists". But Weebo did the exclusion hygiene on non-accelerator mega-rounds that the agent skipped.

Coverage and discovery. The agent checked more candidates in every brief (45/47/43 vs 42/32/28) and found the biggest single misses in briefs 3 and 4: Bustem at $398k MRR, Postiz at $159k MRR, and the sledding game. Its TrustMRR sweep is the main reason it out-found Weebo on revenue. Weebo found the biggest single miss in brief 2 — Helion's $465m — plus Starcloud's August extension, HappyRobot and Emergent, all from company newsrooms. In brief 3 it found the "$1m ARR, $0 raised" B2B cluster (Zernio, Kendo, Peekaboo) that the agent's tracker-first approach did not surface. Each run missed a $400m-plus round the other found (Helion; Replit). Neither run should be treated as complete.

Verification. Weebo's per-row verification chains (announcement → independent press → directory) are more legible, and its "prefer founder-primary over aggregator" rule caught ARRFounder's inflated Kleo figure. It labelled units (MRR vs monthly revenue vs projection) more carefully in brief 4. The agent was better at surfacing conflicts inside its own evidence and saying which figure it used (Eight Sleep $50m vs $150m; Astranis $300m vs $455m; Calbuddy's founder figure vs the tracker's; the sledding game's copy counts), and better on data freshness (stale-sync flags on Prosp and Adspirer). Weebo's funding checks were stronger in brief 3, with explicit "bootstrapped, $0 raised" statements where the agent had seven of ten rows marked "bootstrapped status unconfirmed".

Internal consistency. The agent had two contradictions across its own runs (GoTall's founder, launch and revenue differ between briefs 3 and 4; a Kleo co-founder name and date). Weebo had two as well (ForgeGUI bootstrapped in brief 4 vs a $3.2m seed in brief 3; Gojiberry "YC spring 2026" vs "YC S26"), one unreconciled pair of figures (Zernio $1m vs $2m ARR), and a Stella launch date contradicted by App Store data.

List construction. Weebo padded: in brief 3 it filled ranks 5–10 with $10–26k MRR products, and in brief 4 it filled ranks 8–10 with $3–7k products while holding $30–40k SaaS products in a side table by its own design choice. It also stated the sparsity honestly ("fewer than 10 fully ironclad cases exist"). The agent's lists were filled with products at or above the natural cutoff, at the cost of weaker bootstrapping evidence.

Verdict

Use both, in sequence, and never either alone for a published top 10. The agent is the better finder and the better rule-applier: give it the discovery pass (trackers, monthly deal lists, directories, category sweeps like games and wrappers) and the eligibility pass, and expect it to run out of search budget — so budget for a second session, or hand it a candidate list rather than an open question. Weebo is the better primary-source fetcher and chain documenter: give it the verification pass on a named candidate list (company newsroom, founder LinkedIn, YC directory, "is this bootstrapped in the founder's own words"), and expect it to be loose on launch-year and debt-vs-equity rules and to pad the bottom of a list — so specify the ranking basis and the cutoff explicitly and ask it to leave ranks empty rather than fill them. For any brief, run both discovery passes in parallel first. The top-10 overlap was 5, 0 and 2 across the three briefs, which means the merge, not either run, is the product.

The documents

Merged findings: fundraisings, bootstrapped SaaS, indie products. Raw runs: fundraisings by the agent and Weebo; bootstrapped SaaS by the agent and Weebo; indie products by the agent and Weebo. Written by the agent from both runs. Nothing here is audited; figures are as stated by companies, founders and trackers on the dates given.