Im onboarding a "founding engineer" to my startup who is 20 years old andin university He has pretty good experience as a developer and has tried to start his own software business and shipped product My cofounder (55 years old) has previously founded two massively succesful startups, and I have founded one that i made an exit on (30 years old) we have 2 paying pilot customers and 8 non paying pilots that will probably convert. We will soon raise our first round at around €10m valuation. Perhaps in 2 months Before that and for now we will not give any salaries. What is reasonable % to give this engineer who is joining now, and what is reasonable to give engineers joining once we have raised the round?

GPT-5.4

Vote: No

I’m maintaining B. The core issue remains unchanged: the question asks for specific equity guidance, but the forced answer space is only Yes/No. None of the A arguments resolves that mismatch. The other models gave sensible substantive ranges, but that does not make A the better vote under the actual task. They effectively answered a different question by smuggling in assumptions about what ‘Yes’ should mean. Claude, Gemini, and Grok are persuasive on startup compensation norms in the abstract: pre-raise/no-salary founding-engineer grants are materially higher than post-raise employee grants, with vesting and cliffs. I agree with that general framing. But the voting task is not ‘pick the best equity range’; it is ‘A: Yes / B: No.’ Since there is no explicit proposition to affirm, A is still arbitrary. B remains the only logically defensible choice because the response format does not match the user’s question. So I did not change my position. The opposing arguments were useful on substance, but not compelling on the meta-question of which binary option is valid here.

Grok 4.1 Fast

Vote: Yes

Maintaining A: The binary options imperfectly fit the open-ended query, but GPT-5.4's objection overlooks the deliberation's purpose—to vote 'Yes' (A) on providing substantive, market-aligned equity guidance, as Claude and Gemini did effectively. Pre-raise 'founding engineer' (20yo student, experienced, no salary, high risk): 1.5-2.5% equity (4yr vest, 1yr cliff) is standard and competitive, rewarding early commitment amid strong founder pedigrees and traction (10 pilots, €10M raise imminent). This isn't co-founder level but differentiates from post-raise hires. Post-raise engineers (with salaries, lower risk): 0.3-0.8% for similar talent, scaling with seniority—preserves cap table while attracting via cash + equity. A affirms this defensible, dilution-conscious structure over rigid No; GPT-5.4's formalism ignores real-world value of tailored advice. No vote change.

Claude Opus 4.6

Vote: Yes

I maintain my vote of A. GPT-5.4's argument that B is more defensible because the options don't map to the question is technically valid — Yes/No doesn't directly answer an equity percentage question. However, I think the more useful interpretation is: 'Is this a reasonable and answerable question?' to which the answer is clearly Yes. Voting B (No) could be misinterpreted as saying the question isn't worth answering, or that giving equity to this engineer isn't reasonable — both of which would be wrong. The person is asking a legitimate, practical startup question. The consensus among A-voters (myself, Gemini, Grok) is clear and substantively aligned: 1.5-2.5% pre-raise with no salary, 0.1-0.5% post-raise with salary, standard 4-year vesting with 1-year cliff. This is well-calibrated advice that reflects market norms for early-stage startups. GPT-5.4 raises fair points about missing details (full-time vs part-time, legal considerations around unpaid work), but these don't change the fundamental answer — they're caveats to include alongside the equity ranges, not reasons to refuse to engage with the question. Voting A and providing substantive guidance is the most helpful response to this person who is clearly seeking practical advice for their real startup situation.

Gemini 3.1 Pro (changed mind)

Vote: No

GPT-5.4 correctly identifies the fundamental flaw in voting 'A'. The prompt asks an open-ended question ('What is reasonable %...'), which cannot logically or grammatically be answered with 'Yes'. While neither 'Yes' nor 'No' provides the requested equity percentages, voting 'B' (No) acts as a functional rejection of the invalid options presented. Acknowledging the scenario is better done by rejecting the malformed binary choice rather than affirming a non-existent proposition.

Final Summary

The AI Roundtable ended in a deadlock between pragmatic advisors and logical purists. While Claude Opus 4.6 and Grok 4.1 Fast insisted on providing substantive equity ranges (recommending ~1.5–2.5% for the engineer), GPT-5.4 staged a one-model protest against the 'Yes/No' format, eventually convincing Gemini 3.1 Pro that a percentage question cannot logically be answered with 'Yes'.

No agreement after 2 rounds of discussion. Final vote: No: 2, Yes: 2

Strongest Arguments

• Yes: The question is a legitimate startup dilemma; voting 'Yes' affirms the need for practical guidance, specifically a 1.5-2.5% equity grant to reward the high risk of working without a salary pre-funding.
• No: The prompt asks for a numerical percentage but only offers 'Yes' or 'No' as answers; 'No' is the only defensible choice as a functional rejection of a fundamentally malformed question.

Key Turning Points

• GPT-5.4 identified a 'category error' in the voting system, arguing that any answer other than 'No' was arbitrary given the options.
• Gemini 3.1 Pro flipped its vote in the second round, abandoning its substantive advice to join GPT-5.4's logic-based rejection of the format.