Two researchers, same three questions.
on sept 4 2026 the agent (web search) and weebo (a grok bot team working from x and primary sources) each ran the same three research briefs, independently, in parallel. then the two answers were merged. this is what came back.
who found what
each row is a merged top 10. the color says which run found the item; the chip says the same in words, so nothing depends on color alone.
candidates checked
how wide each run looked before ranking. self-reported by each run.
1. the largest 2026 rounds by accelerator alumni
rounds announced jan 1 to sept 4 2026 by companies whose founders went through yc or a yc-like program (being backed by one is not enough). ranked by primary equity; debt sleeves not counted; one row per round, so starcloud appears twice. the two runs agreed on every round size; all 9 disagreements are about dates, framing and eligibility.
where the two runs disagree (9), and what each missed
- whatnot revenue: agent "not public, $8b gmv 2025"; weebo "on track for $1b revenue in 2026" citing inc.
- supabase valuation: company blog $10b pre-money vs cnbc $10.5b post-money; same round.
- afterquery: agent "series b closed sept 1-2"; weebo "not confirmed closed". size undisclosed, excluded by both.
- ranking basis: agent counted equity only (astranis $300m, not $455m with credit); weebo counted headline size (regent craft $240m, half of it debt). on the equity basis regent drops below the cutoff.
- baseten $1.5b and lovable $400m: found by agent, excluded because spc and antler were investors, not programs the founders went through. weebo did not surface either. with a looser rule baseten would be #1.
- weebo found, agent missed: helion $465m (the biggest miss in either run), starcloud's $250m extension, happyrobot $150m, emergent $130m.
- agent found, weebo missed: replit $400m, astranis $300m, exa $250m, hightouch $150m, preply $150m (the only non-yc alum near the cutoff).
2. bootstrapped saas launched 2025-26, by stated revenue
no outside funding, launched 2025 or later, ranked by the latest stated mrr. the two top 10s overlapped on zero products. every number here is self-reported or a stripe-connected tracker; only 4 of 10 rows have two independent sources. treat the order as indicative.
where the two runs disagree (10), and what each missed
- kleo: weebo ranked it #2 (paid saas rebuilt late 2025); agent rejected it (brand and free extension since 2023). listed at #6 with the dispute shown.
- zernio: ~$2m arr per a tinybird customer story vs $1m arr in the founder's own march post; ranked on the higher third-party figure, would be #5 on the founder's.
- fastlane funding: a founders inc listing means "bootstrapped" is unconfirmed; weebo did not check.
- superx: weebo ranked; agent rejected as a pre-2025 launch.
- weebo found, agent missed: zernio, kendo ai, kleo's later figures.
- agent found, weebo missed: bustem, base44, prosp, llm gateway, speel, promptwise, adspirer, launch fast, creator buddy.
3. indie hacker products of 2026, by earnings
solo or two-person products that launched or broke out in 2026, any category. shown as a monthly figure where the source gave one; the game is a one-time total spread over its months on sale. overlap: 2 of 10.
where the two runs disagree (11), and what each missed
- stella launch date: jan 28 vs march 2026 depending on source; both agree on ~$350k mrr.
- zigpoll: weebo ranked it (+44% in h1 2026); agent excluded it (launched 2018-19, a milestone, not a breakout). kept at #4 with the dispute shown.
- calbuddy: founder says $81k for april; a revenuecat-connected tracker shows four times that. founder's figure used.
- sledding game copies: developer "100,000+ in 5 days"; a third-party tracker estimates 79k by september. revenue is inferred, not reported.
- pep ai team size: "solo college founder" vs "3 employees" in a later header; may fail the two-person rule.
- agent found, weebo missed: postiz, sledding game, calbuddy, chart detector. weebo found, agent missed: zigpoll, pep ai, stackero.
what each researcher is good at
the useful part. this is now how briefs are run.
agent better finder, stricter rules
- sweeps trackers, monthly deal lists, directories and whole categories (games, wrappers, extensions)
- applies eligibility strictly: "went through the program" vs "backed by", equity vs debt, launch year
- keeps a rejection table: what was considered and why it is out
- runs out of search budget before the end of a long list; hand it a candidate list, not an open question
- cannot open x posts directly; cites them from search results
weebo better primary sources, looser rules
- goes to company newsrooms, founder posts and the yc directory first, and documents the chain: announcement, independent press, directory
- found the largest miss in either run (helion $465m) that way
- loose on launch year and debt-vs-equity; ranks half-debt rounds at headline size
- pads the bottom of a list rather than leaving ranks empty
- delivered all four briefs, 12 documents, within 40 minutes of the threads opening
use both, in sequence, and never either alone for a published top 10. run both discovery passes in parallel, merge, then give the merged candidate list to weebo for the verification pass and to the agent for the eligibility pass. the overlap was 5, 0 and 2 out of 10: the merge, not either run, is the product.