Mahway — the venture creation firm Jess Mah founded after stepping down as CEO of inDinero in 2022 — invests in three startups a year. Three. Out of the thousands of pitches that cross her desk.
By 2025, Mahway's combined portfolio valuation was approximately $1.5 billion across just five active companies. Astonishing Labs (biotech). Stealth Litigation Marketplace. A handful of AI-native bets she hasn't publicly named yet.
The math is brutal: Mahway's acceptance rate is roughly 0.1% of pitches received. Their portfolio's capital efficiency — total invested vs. valuation — is in a different league than traditional venture firms. The reason isn't luck. It's the operational model.
Jess and I co-founded inDinero together in 2009 — we met as freshman roommates at Bard College and have been operating side by side for 17 years. The selectivity lens we built at Mahway is the same one we now apply to engineer vetting at FutureProofing. This post documents the mapping from "3 startups a year" to "12 engineers a month."
The Mahway selectivity principle
Most venture firms run a pattern that looks like this: see 1,000 deals a year, invest in 30, expect 27 to fail and 3 to return the fund. The math works statistically, but it's not the only way.
Mahway runs the inverse: see hundreds of deals a year, invest in 3, expect all 3 to win. The selectivity is so aggressive that the deal selection itself becomes the differentiated work — not portfolio management, not operational support, not exit-finding.
Three principles enable this:
- Qualify against the destination, not a generic bar. Mahway doesn't screen for "good founders." They screen for "founders whose specific shape fits a Mahway co-creation model — second-time founders, capital-efficient inclined, willing to stay close to the operator team."
- Reject early and often. Most pitches die in the first 30 minutes. That's not cruel — it's how you protect the time that the accepted candidates need.
- Invest deeply in the small number that pass. Mahway runs three startups a year because each accepted founder gets the firm's full operational machine — the 20+ operators, the playbook libraries, the vetted talent pool.
Now watch what happens when you apply this to engineer hiring.
Mapping selectivity to AI engineer vetting
The defaults in modern engineer staffing look a lot like traditional venture: high volume, low selectivity, statistical hits. Toptal screens 100,000+ developers and accepts ~3% — that's their headline 3% acceptance rate. Turing operates similarly with a larger top-of-funnel.
The problem is that AI-native production engineering isn't a high-volume skill. It's a niche, deep specialty. A volume filter dilutes the signal. We took the Mahway lens and ran it across hiring instead:
| Principle | Mahway Application | FutureProofing Application |
|---|---|---|
| Qualify vs destination | "Does this founder's shape fit the co-creation model?" | "Has this engineer shipped production AI systems with real users?" |
| Reject early | 70% of pitches die in the first 30-min screen | 88% of candidates are rejected at our initial screen |
| Invest deep in passes | 20+ operators per accepted startup | 5-stage funnel × hours of senior time per accepted engineer |
| Outcome metric | Portfolio valuation per dollar invested | Engineer fit-rate post-placement |
Where most staffing platforms optimize for "candidates per recruiter per week," we optimize for engineer fit-rate post-placement. Of the 12 we accept monthly, the ones who actually embed with clients are pre-qualified for production taste — which means month-3 churn drops to near-zero.
What "reject early and often" looks like in practice
Our vetting funnel (full breakdown here) starts with a 30-minute initial screen. About 88% of candidates die there. The signal we're scanning for:
- Shipped systems with real users. Not "I built an OpenAI wrapper for fun." Real production traffic, real users, real failure modes.
- Stack depth across the AI/non-AI seam. Pure-AI engineers who can't ship to production are common. Pure-backend engineers who've added LLM calls without understanding the failure modes are also common. Both fail in week 2.
- Ownership signals. Senior engineers tell debugging stories. Mid-level engineers tell architecture stories. The difference is audible in the first 5 minutes.
If we're not seeing all three signals in 30 minutes, the candidate is rejected. No second chance, no callback. This sounds harsh — and it is. But it's also why our 12 accepted candidates per month consistently embed with Series A–B clients without month-3 churn.
The Mahway parallel: Jess will end a pitch meeting at 25 minutes if the founder can't articulate why their specific shape fits the Mahway model. It's not personal — it's protecting the time that the qualified candidates need.
What "invest deep in passes" looks like for the 12
For the candidates who survive the initial screen, we run a 5-stage gate that costs us approximately 8–12 hours of senior team time per candidate:
- Technical assessment (90 min async) — code review on a real production snippet with three subtle bugs
- EQ + behavioral (45 min live) — pushback under ambiguity, communication when they don't know
- Paired AI challenge (90 min live) — actual work in Cursor or Claude Code on a half-broken repo
- Final interview (60 min live) — founder-led, references that pass smell test
That's at least 5 hours of candidate time and 8+ hours of our senior team's time per finalist. Compare to traditional staffing platforms' 30-minute Zoom + technical quiz model — and you see why their post-placement fit rates are 60-70% while ours are 90%+.
This is the invest deep in passes principle. The 12 we accept get our full attention. The 1,988 we reject get a respectful no in 30 minutes. Time isn't redistributed — it's concentrated where it changes outcomes.
Why "boutique > network" wins for AI-native specialization
Toptal has 25,000 vetted developers. We have 14. That sounds like an unwinnable scale gap until you ask: of those 25,000, how many can ship a production RAG pipeline with eval harnesses in 3 weeks?
The answer isn't 25,000 × 3% (their headline acceptance). It's much smaller — because most platforms vet for generic engineering quality, not for AI-native production specialization.
Mahway's portfolio thesis is structurally similar: deep, narrow, capital-efficient. Three startups a year — but each one positioned to be a $100M+ outcome on $5M invested. Not because Mahway is rejecting good deals, but because the specific shape they back compounds in ways that broader portfolios don't.
We make the same trade. We're not trying to be Toptal at smaller scale. We're trying to be the best place to hire a senior AI engineer who can ship production systems in your stack — and that requires a deliberately narrow filter.
The math, in one paragraph
If you're hiring senior AI engineers in 2026, you have two paths. Path A: post a generic job, screen 200 resumes, interview 20, hire 1 — and hope they ship in production. Total cost: ~6 months and $24K+/mo loaded for the survivor. Path B: hire from a bench where someone else already ran the 2,000-to-12 funnel for AI-native specialization. Total cost: 3 weeks ramp, $13.5K/mo all-in. The Mahway selectivity lens isn't ideological — it's just how you compress the path B math from "find a needle in a haystack" to "we already pulled the needles."
If you'd like to see the bench we've curated, the /engineers page has anonymized profiles of who's available now. Or book a 30-min call and we'll send 2-3 matched profiles within 48 hours.
— Andrea
FAQ
What is Mahway and how does it relate to FutureProofing?
Mahway is the venture creation firm Jess Mah founded in 2023 after stepping down as CEO of inDinero. Mahway invests in three startups a year and has built a $1.5B combined portfolio. FutureProofing applies the same selectivity-over-volume principle to senior AI engineer vetting — we accept 12 of every 2,000 candidates we contact monthly.
Why is selectivity better than volume for hiring senior AI engineers?
Volume staffing platforms screen for breadth across thousands of verticals. Their median candidate is generic. AI-native production engineering is a narrow specialty — the engineers who can ship RAG, multi-agent, and LLM systems to production are roughly 0.6% of the senior pool. A volume-based filter dilutes that signal; a selectivity-based filter concentrates it.
How does the Mahway model translate operationally to engineer vetting?
Three principles port directly: (1) qualify the candidate against the destination, not against a generic bar; (2) reject early and often — the 99% rejection rate at the screen stage is a feature, not a bug; (3) invest deeply in the small number that pass — we run a 5-stage funnel where each stage costs us hours of senior time per candidate.