# Three weeks after the YC Summer 2026 RFS, the Spring batch is the field check

URL: https://www.thedeepfeed.ai/posts/2026-05-22-yc-spring-2026-rfs-three-week-field-check/
Category: Business
Published: 2026-05-20
Updated: 2026-05-22
Author: the-deep-feed
Tags: y-combinator, requests-for-startups, spring-2026, openai, founder-strategy
Kind: deep

> The Spring 2026 batch is the empirical answer to the April 30 RFS read. Eight predictions held, four broke, and OpenAI just rewrote every cap table.

## TL;DR

- The [April 30 RFS read](https://www.thedeepfeed.ai/posts/2026-04-30-yc-rfs-summer-2026-decoded/) made testable predictions. The **Spring 2026 batch (171 companies, 55 publicly launched)** is the field check.
- **The 'horizontal Company Brain is dead' prediction broke.** Hyper, Ontora, Memory Store and Wato all pitch horizontal company-knowledge layers. YC funded all four.
- **Software for Agents took the largest single share, as predicted.** 16 of 55 visible companies sit in that one lane. Three of four asymmetric silent lanes stayed silent; Counter-Swarm Defense broke on **Arlo Industries**.
- Applying [Sequoia's services-as-software checklist](https://www.thedeepfeed.ai/posts/2026-04-30-sequoia-services-as-software-thesis/) to the 16 RFS lanes: **4 lanes fit the $1T shape cleanly**, 3 are hybrids, **9 are not the Bek shape**. Plus **12 specific Bek-shape pitches** sized to fit the **May 25 Summer 2026 deadline**.
- OpenAI's [$2M-tokens-for-uncapped-SAFE offer (May 20)](https://www.thedeepfeed.ai/posts/2026-05-22-openai-2m-tokens-yc-batch-equity/) made every Spring 2026 company a tokenmaxxing experiment on the same instrument.

Three weeks ago, [The Deep Feed published a reading of the YC Summer 2026 Request for Startups](https://www.thedeepfeed.ai/posts/2026-04-30-yc-rfs-summer-2026-decoded/) that staked its argument on the chatter, not the list. The thesis was that launch-week engagement, OSS saturation and partner amplification together painted a pre-mortem for the cohort that would actually get funded. The piece made eight specific predictions and three more conditional ones. Most of them were testable on a horizon of weeks, not months.

The Spring 2026 batch is the horizon. YC reports 171 companies in the batch. Fifty-five have publicly launched their company pages as of today. That is roughly 32% of the cohort — enough to do an empirical read, not enough to call the verdict final. And one additional variable has landed on the same day this piece is going out: [OpenAI's $2M-tokens-for-equity offer to every Spring and Summer 2026 company](https://www.thedeepfeed.ai/posts/2026-05-22-openai-2m-tokens-yc-batch-equity/), structured as an uncapped SAFE that converts at the next priced round.

This piece does one job. It scores the April 30 read against what the 55 visible Spring 2026 launch pages actually say, and it folds in the OpenAI deal as the new structural variable on top of the original framework. The discipline is field-check, not editorial. Where the predictions held, the reasoning is named. Where they broke, the contradiction is named first and the explanation second.

{/* IMG-PROMPT: held-vs-broken-scorecard: A simple two-column editorial scorecard on cream paper. Left column labeled "HELD", right column "BROKE". Each column has small abstract icons stacked vertically — checkmarks on the left side, X marks on the right — with the left column slightly taller than the right. One small editorial red accent on the top of the right column. Style: editorial illustration, cream paper background #f6f1e7, charcoal ink #1a1612, single editorial red accent #e63946, NO screenshots, NO text labels visible (or only one or two short ones), NO realistic logos, hand-drawn editorial quality. */}
![A two-column scorecard. Eight predictions held. Four broke. The held column is taller. The broken column is the editorial reading.](/post-images/2026-05-22-yc-spring-2026-rfs-three-week-field-check/held-vs-broken-scorecard.jpg)

# The visible 55 vs the missing 116

The methodology caveat is load-bearing. The YC Spring 2026 batch listing reports 171 companies. The launched-companies filter renders fewer than half of them, with the page itself noting that more will be added as Spring 2026 launches continue. After de-duplicating the launched listing against the B2B-filtered listing and pulling the Ontora detail page, the addressable visible cohort is 55 of 171, or 32%. The remaining 116 companies have not yet released public pages and are not in the source set.

That coverage gap is the first reason to keep the verdict directional. The visible 55 is not a random sample. It is the slice of the batch whose founders have already moved their pages out of stealth, which means it skews toward companies whose pitches are fully formed enough to ship a public one-liner. A horizontal "Company Brain" play is, on average, easier to write a one-liner for than a vertical legal-data ingest play that requires the partner to take twenty minutes of context. So the visible-batch sample probably over-represents the easy-to-pitch horizontals and under-represents the slow-to-pitch verticals. That bias matters for one of the broken predictions below.

Within the 55, 39 companies have a one-liner that maps cleanly to one of the 16 RFS items. Four of those 39 cross-fit two lanes (Memory Store, Wato, Tasklet and Hessian each sit at the boundary of two items). Sixteen of the 55 sit outside the RFS frame entirely: consumer products, generic robotics, gaming, fintech derivatives and creator tooling. YC funded those 16 in the same batch as the 39 RFS-aligned companies, which is the first reminder that an RFS is a wish list, not a hiring rubric.

The lane distribution inside the 39 is uneven in the way the April 30 piece predicted. Software for Agents is the single largest lane at 14 unambiguous claimers plus two cross-listed, for 16 total. AI-Native Service Companies is second at 10. Company Brain has four. The AI Operating System for Companies has three. Five of the sixteen lanes have zero visible claimers: AI for Low-Pesticide Agriculture, Electronics in Space, Industrial Capabilities in Space, Inference Chips for Agent Workflows, and Supply Chain 2.0 for Semiconductors. That zero-claimer set is the second-most-important number in this piece, after the 16 in Software for Agents, because those five lanes are exactly the asymmetric silent lanes the April 30 read tagged as the highest-EV slots in the batch.

# What held up

The April 30 piece made eight specific predictions and two conditional ones. Walked one by one against the visible Spring 2026 batch, the score is meaningfully positive.

🟢 **"Software for Agents will take the largest single share of the S26 cohort."** Confirmed, by a wide margin. The lane holds 16 of 55 visible companies — roughly 29% of everything publicly launched in the batch. Five names give a feel for the saturation: AgentPhone ("the phone for AI agents"), Indexable (a sandboxed execution layer), Pentagon (agent orchestration), OpenWork ("the open source alternative to Claude Cowork"), and Klaimee ("liability insurance for AI Agents"). Inside that 16, the dominant pattern is exactly what the April 30 read predicted: specific primitives rather than generic frameworks. Agent auth (Clawvisor), agent insurance (Klaimee), agent sandboxes (Ardent, Indexable, Runtime), prompt-injection defense (Silmaril), agent documentation (Manicule), agent staging (Chronicle Labs). The handful of generic-framework pitches in the lane (Pentagon, OpenProse) are the minority, not the median.

🟢 **"Industrial Capabilities in Space — effectively zero public claimers."** Confirmed. Zero P26 company in the visible 55 mentions regolith, lunar extraction, in-space manufacturing, electrolysis, or 3D printing in space. The lane is empty in the exact way Adi Oltean's launch-week tweet implied it would be: founders in this category apply through DMs, not RFS replies. If YC funds a team here, it will not be visible in the launched-companies listing for another month.

🟢 **"Electronics in Space — effectively zero public claimers."** Confirmed. Zero P26 company in the visible 55 has a one-liner combining "space" with "satellite", "orbit", "rad-hard" or "inference chip". Philip Johnston's lane mirrors Oltean's. The founder pool is small, technical and off X.

🟢 **"AI for Low-Pesticide Agriculture — effectively zero public claimers."** Confirmed. No P26 company's one-liner contains "agriculture", "pesticide", "weed detection", "precision spraying", "microbe biologicals" or "RNA agricultural inputs". Garry Tan's only RFS-touching tweet of launch week pulled 96 R+Q in the original engagement table; the founder population it pulled has not yet shown up in the visible launched companies. If Tan funds one team here, it will not be from anyone who tweeted at him in launch week.

🟢 **"Supply Chain 2.0 for Semiconductors — effectively zero public claimers."** Confirmed. The lane that had the lowest launch-week engagement of the 15 items (7 R+Q) is also the lane with zero visible claimants three weeks later. The April 30 explanation that this is the quietest possible founder population stands. No P26 one-liner contains "TSMC", "HBM", "advanced packaging", "wafer allocation", "fab supply chain" or "export-controls tooling".

🟢 **"Inference Chips for Agent Workflows is high engagement, low claiming."** Confirmed. The lane pulled 102 R+Q in the launch table (the third-loudest item by founder engagement) and produced zero visible Spring 2026 claimers. The closest one-liner in the visible batch is Expanse ("Unlock wasted GPU capacity"), which is a GPU resale and scheduling play, not a silicon company. The April 30 framing was that capital-intensity is the moat and the founder pool for inference silicon is small and applies privately. That explains the absence cleanly.

🟢 **"The OSS-saturated lanes will produce the highest applicant volume."** Confirmed. Software for Agents (with mem0 at 54k stars, Composio at 28k, awesome-mcp-servers at 86k, A2A at 23k) and AI-Native Service Companies (low OSS density, high engagement) are the #1 and #2 lanes by visible-batch claimer count. The argument the April 30 piece made was that a 50,000-star incumbent does not deter founders. It attracts them, because founders read public OSS density as evidence of demand. That is exactly what the Spring 2026 batch shows. The partner reading the Software for Agents applications is going to be the most skeptical reader in the room. That was the April 30 prediction. The 16-team density inside the lane is the empirical answer.

🟡 **"Asymmetric items will produce one or zero teams each."** Partial. Four of the five asymmetric items (Industrial Space, Electronics Space, Supply Chain 2.0, Low-Pesticide Ag) hold at zero. Counter-Swarm Defense broke at exactly one: Arlo Industries, the passive aerial sensing mesh play, hits the lane directly. The strong-form version of the prediction ("effectively zero") broke on Bosmeny's item. The "one or zero per asymmetric lane" budget, taken as a budget, held. Arlo is precisely the one team the rule allowed.

That is eight predictions held in some form out of ten tracked at the 3-week mark, before the broken predictions are addressed. The structure of the April 30 read, namely that founder engagement on launch day predicts visible-batch claimer count three weeks later, with the asymmetric lanes producing one-or-zero teams each, survived contact with the data.

The two predictions that cannot be scored at 3 weeks are the ones that need 18-month attrition data: "the most-funded category will have the highest mortality" and "the AI-Native Service Companies item will produce one breakout and a lot of $1M-ARR plateaus." Neither is testable in May 2026. Both become testable in late 2027.

{/* IMG-PROMPT: software-for-agents-stack: A vertical editorial stack of sixteen identical horizontal bars on cream paper, like books stacked in a tower or layers of a strata diagram, slightly leaning. The top three bars accented in editorial red. The image conveys density and saturation in a single lane. Style: editorial illustration, cream paper background #f6f1e7, charcoal ink #1a1612, single editorial red accent #e63946, NO screenshots, NO text labels visible (or only one or two short ones), NO realistic logos, hand-drawn editorial quality. */}
![A stack of sixteen layers. The Software for Agents lane in the Spring 2026 batch — roughly one-third of every visible company.](/post-images/2026-05-22-yc-spring-2026-rfs-three-week-field-check/software-for-agents-stack.jpg)

# What broke

The April 30 read had three predictions that broke on the visible batch. One of them is the editorial centerpiece, and it is the one worth opening with: the horizontal Company Brain pitch did not die. It got funded four times.

🔴 **"The horizontal Q&A bot is dead because Notion exists; the wedge that survives is vertical."** This was the load-bearing call inside the Company Brain section of the April 30 piece. The argument turned on Zach Tratar's three-word reply to Tom Blomfield's launch tweet ("Have you tried Notion?"), which pulled 219 likes and 88,279 impressions and no public response from any YC partner. The reasoning was that Tratar's bona fides on the problem (ex-founder of Embra, acquired by Notion in 2025, now working on Notion AI) made the question impossible for a partner to ignore. The April 30 prediction was that any founder pitching a horizontal Q&A layer over company knowledge would be filtered out by the Tratar question, leaving vertical Company Brain plays (legal, healthcare, regulated-data) as the only surviving wedge.

The Spring 2026 batch funded four Company Brain pitches. None of them is vertical. All four are horizontal.

**Hyper** has the most literal RFS-restatement one-liner in the entire visible batch: "The Self-Driving Company Brain." Ontora describes itself as "AI agents that interview every employee to map how work gets done," which is Blomfield's "tribal-knowledge compiler" framing rendered into a product. Memory Store sells "shared context for all of your team's agents." Wato sells "shared memory, tools, and workflows for agents across teams." Memory Store and Wato cross-fit Software for Agents, but the framing in both one-liners (shared memory across a company's agent fleet) is the horizontal Company Brain play in a slightly different vocabulary. None of the four names a single vertical. None of them mentions legal, healthcare, financial services, regulated-data ingestion, or any other defensible niche.

The Tratar question, as a filter, did not work. YC funded the exact pitch the April 30 piece said partners would discount, and the most direct version of that pitch (Hyper's one-liner) is one of the most pure horizontal-restatement one-liners in the batch. There are two charitable readings of the contradiction and one uncharitable one. Charitable: the visible-batch slice (32%) under-represents vertical Company Brain plays, which are harder to write a launch one-liner for, and the vertical wedges are in the unlaunched 116 companies. Charitable: Notion AI is genuinely worse than the April 30 read assumed, leaving more horizontal addressable surface than Tratar's tweet implied. Uncharitable: the April 30 piece over-indexed on a single skeptic's tweet and the partners reading the applications discounted Tratar's framing more than the original analysis predicted. The empirical answer at three weeks is that four horizontal Company Brain pitches walked in, and four walked out funded.

🔴 **"Diana Hu's three RFS items are all silent."** This was the framing the April 30 piece used for Diana Hu's posture as the quietest of the partner amplifiers. The prediction was that her three items (Inference Chips for Agent Workflows, The AI Operating System for Companies, and Supply Chain 2.0 for Semiconductors) would produce few public claimants because founders in those categories apply through quieter channels.

Two of the three held. Inference Chips has zero visible claimers. Supply Chain 2.0 has zero visible claimers. The AI Operating System lane broke, and it broke in three places. Modern positions itself as "the AI-native OS for IT," ProjectX is "Infinity: the first OS where humans and agents work with no limits," and Tasklet is "the cloud agent operating system for knowledge work." Three OS-shaped claimers, one of them (Modern) the cleanest possible vertical fit for Hu's "queryable company" thesis. The blanket "Hu silent" framing was wrong on the OS item specifically. The blanket framing was also too coarse. The right read three weeks later is that *two* of Hu's three items are silent, not all three, and the AI Operating System item is the lane where Hu's posture and the founder population diverge.

🔴 **"Counter-Swarm Defense will have effectively zero public claimers."** This was filed under the asymmetric-silent-lanes bucket alongside the four other quiet items. The strong-form wording broke on Arlo Industries. The lane is not zero; it is one. The one-or-zero asymmetric budget, treated as a budget rather than a hard zero, still survives (Arlo is precisely one team), but the specific "effectively zero" phrasing the April 30 piece used does not. The offensive mirror at Maquoketa Research, which ships *"one-way attack drones with onboard targeting"* per its YC page, sits outside Counter-Swarm proper but in the same defense-tech adjacency, suggesting the founder population in this category is less off-X than the April 30 read assumed.

A fourth prediction sits in a grey zone. The April 30 piece argued that "the horizontal generic 'tooling for agents' pitch is dead; only the narrow wedges (auth, billing, MCP registry) survive." The 14-plus-two visible Software for Agents claimers are mostly specific primitives: Clawvisor on auth, Klaimee on insurance, Silmaril on prompt-injection defense, Manicule on docs, Chronicle Labs on staging, Indexable and Runtime on sandboxes. That is the prediction holding. But a minority of the lane (OpenProse, Pentagon) did get funded on horizontal framing-only pitches, which is the prediction breaking softly. The right grade is "mostly held, with two exceptions that suggest the partner bar on a generic horizontal SfA pitch is lower than the April 30 read claimed."

Net tally at three weeks across the visible 32% of the batch: eight predictions held, three cleanly broke, one mixed, and two cannot be scored yet. The broken predictions are concentrated in one structural place. The April 30 read assumed that horizontal pitches in contested lanes would not clear the partner bar, and that single structural mistake explains both the Company Brain break and the AI Operating System break. The original analysis under-estimated YC's tolerance for horizontal pitches in lanes where the OSS competitors are real and well-funded.

# Six companies worth watching

The Top 12 most editorially interesting names from the visible batch include some that are interesting because they hit an RFS lane verbatim, and some that are interesting because they hit a wedge the April 30 piece flagged as missing. The six below are the ones whose pitches sharpen the field check directly.

[**Arlo Industries**](https://www.ycombinator.com/companies/arlo-industries). Counter-Swarm Defense. Tyler Bosmeny's lane.

> Passive aerial sensing mesh to track drones and missiles.

Arlo is the single broken silence inside the asymmetric set. The April 30 read called the Counter-Swarm lane "effectively zero" and budgeted one team across the four asymmetric items combined; Arlo is exactly one. The one-liner is also the most direct possible hit on Bosmeny's RFS framing of "software that fuses every sensor and every defender on a site into a single real-time picture." Passive sensing mesh is the input layer of that fusion. Bosmeny's own posture three weeks ago was that defense-tech founders run stealth and apply privately. Arlo's public launch is the first signal that at least some of the defense-adjacent founders in the Spring 2026 batch were willing to flip the stealth posture early. If Bosmeny reads only what is public three weeks in, Arlo is, by default, his reference implementation for the lane.

[**Hyper**](https://www.ycombinator.com/companies/hyper). Company Brain. Tom Blomfield's lane.

> The Self-Driving Company Brain.

Hyper is the most verbatim RFS-restatement one-liner in the visible batch. Blomfield's launch tweet framed Company Brain as the executable, agent-shaped knowledge layer for every company; Hyper's one-liner is that sentence compressed into five words. It is also the cleanest possible contradiction of the April 30 reading. The Tratar question ("Have you tried Notion?") did not stop Hyper from getting funded. The horizontal pitch the original analysis called dead is, in this case, the most polished version of the pitch in the lane. The founder team's bet is that "self-driving," meaning the autonomy promise, is the wedge that survives against Notion AI, on the argument that Notion's surface is still mostly retrieval and not autonomous workflow. That is a real wedge if "self-driving" is operationally true and not just a slogan. The thirty-month read on Hyper is whether they ship something Notion AI cannot, before Notion AI ships something Hyper cannot.

[**Klaimee**](https://www.ycombinator.com/companies/klaimee). Software for Agents. Aaron Epstein's lane.

> Liability insurance for AI Agents. You deploy agents, we cover you.

Klaimee is the most quotable one-liner in the entire batch, and it is the wedge the April 30 piece named as missing under Software for Agents. The original read flagged a list of narrow primitives the SfA lane needed but did not yet have: agent auth, agent billing, agent docs, agent reputation registries, agent insurance. Klaimee is the insurance slot, filled. The Mosaic-shaped business behind the one-liner is the interesting question. Actuarial pricing on agent failure modes is genuinely new, and the canonical comparison (cyber insurance) took fifteen years to develop reliable loss curves. If Klaimee can underwrite agent liability on twelve months of data the way cyber underwriters now underwrite on incident-response data, the company is the first Series A in this corner of the agents stack. If it cannot, the pitch is a wedge without a moat. Either way, it is the precise primitive the April 30 piece said the SfA lane was missing, and YC funded it.

[**Modern**](https://www.ycombinator.com/companies/modern). The AI Operating System for Companies. Diana Hu's lane.

> The AI-native OS for IT.

Modern is the cleanest contradiction of the "Hu items are silent" prediction. The launch-week reasoning was that Hu's posture (quietest of the partner amplifiers) implied her three lanes would pull founders who apply outside the public chatter; the visible-batch evidence is that one of the three lanes (AI OS) pulled three public claimers, and Modern is the vertical-IT wedge among them. Vertical matters here in a way it did not matter for Hyper: the AI Operating System lane is structurally a horizontal pitch that needs a vertical anchor to clear the "queryable company" bar Hu's RFS describes. ProjectX is fully horizontal ("the first OS where humans and agents work with no limits"). Tasklet is fully horizontal ("the cloud agent operating system for knowledge work"). Modern is the only one of the three with a named vertical wedge (IT departments), and that wedge is the surface that ServiceNow and Atlassian's Jira service-desk built billion-dollar businesses on top of. The right read on Modern is that it is the version of the April 30 Company-Brain-vertical prediction that the Company Brain lane itself failed to produce.

[**Panacea**](https://www.ycombinator.com/companies/panacea). AI-Native Service Companies. Gustaf Alströmer's lane.

> AI-Native FDA Regulatory Services.

Panacea is the most precise Gustaf Alströmer alignment in the batch. The launch-week tweet for AI-Native Service Companies pulled 854 likes and 95 R+Q (top-five by founder engagement), and the lane is now the second-largest in the visible batch at 10 claimers. The April 30 pre-mortem on this lane was specific: it predicted one breakout and a lot of $1M-ARR plateaus, with the exit from the plateau being ownership of a regulated-license stack. Panacea is the one-liner that walks straight at the regulated-license thesis. FDA registrations are a license-shaped surface area (IND filings, 510(k) submissions, drug-master files), and a service company that owns the workflow under those filings has the regulated-license-stack moat the April 30 piece described. Arctic Health (payor credentialing) and Huscarl (actuarial certifications) are in the same shape; Panacea is the cleanest. If Alströmer is right that the breakout in this lane will be "the team that walks out with the license, not the team that walks out with the demo," Panacea is the highest-prior team in the visible batch on that bet.

[**Astraea**](https://www.ycombinator.com/companies/astraea). AI Personalized Medicine. Ankit Gupta's lane.

> Agents that accelerate clinical trials.

Astraea is the one explicit claimer in Gupta's lane, and it sits at the intersection of agents-as-product and medicine. The April 30 read of the AI Personalized Medicine lane was that medical founders run quiet and that the strongest pitches in the lane would apply privately. The launch-week numbers (252 likes, 24 R+Q, bottom-quartile engagement) supported that. Astraea is the visible-batch counterevidence: a clinical-trials-agent company that did launch publicly. But the one-liner is closer to Alströmer's AI-Native Services frame than to Gupta's n-of-1 personalized-therapy frame. There is no genome, no mRNA, no wearable-signal language in the pitch. The true n-of-1 personalized-medicine wedge (FDA-pathway-aware therapy design, individualized RNA, on-demand diagnostics) is unclaimed in the visible 55. Astraea is the visible front door of Gupta's lane; the bulk of the lane is still where Gupta said it would be, behind the door.

Three companies from the Top 12 are worth flagging in shorter form for completeness. **Asterlab** ("the first AI-native AI research lab") is the sole claimer in Jon Xu's quiet AI-Native Discovery Engines lane, and it is the boldest single positioning statement in the batch. Saudara AI ("AI sourcing broker linking brands to top global factories") is the only Spring 2026 company in Nicolas Dessaigne's Hardware Supply Chain lane, and it sits on the procurement side of his frame rather than the rapid-iteration prototyping side he flagged in launch week. Silmaril ("prompt injection defense that is self-healing") is the security primitive under Software for Agents that the April 30 piece named as a real wedge, and it is one of the rare SfA pitches with a specific failure mode (prompt injection) in the one-liner rather than handwaved abstractions about "agent infrastructure."

# The OpenAI $2M variable

The field check above was written against the batch as it stood. Three days ago, the batch's structural conditions changed. On May 20, 2026, OpenAI offered $2M in API credits to every company in YC's Spring 2026 and Summer 2026 batches in exchange for equity, with the instrument an uncapped SAFE that converts at the next priced round. [The Deep Feed covered the deal mechanics in a separate piece today](https://www.thedeepfeed.ai/posts/2026-05-22-openai-2m-tokens-yc-batch-equity/); the version that matters for this field check is the cap-table version.

The framing came from Sam Altman the night the offer was announced at a YC event.

> i am excited to see what will happen with tokenmaxxing startups, both for how they work internally and the products they can build. openai offered to invest $2M in tokens into every startup in the current yc batch. happy building!
>
> — [@sama](https://x.com/sama/status/2056933166875857290), May 20, 2026

OpenAI's Rohan Varma published the comparative-economics version of the deal the next day, putting an explicit price on tokenmaxxing against Anthropic.

> Yesterday @sama just offered to invest $2M of OAI credits into any current YC company as an uncapped SAFE. You can use the tokens on the API. This means you can offer agentic products to customers without worrying about price or charging while you find PMF. OAI tokens are worth 2X Anthropic tokens. Our frontier models are ~50% more token efficient than Ant's. This means our $2M in GPT tokens is worth $4M in work accomplished with Opus.
>
> — [@TheRohanVarma](https://x.com/TheRohanVarma/status/2057648423873270270), May 21, 2026

A YC P26 founder, Anand Paj, posted the founder-side read directly. The phrasing is the line that has stuck in the founder chatter.

> I'm part of the YC P26 batch that got offered $2M by Sam Altman in funding with OpenAI tokens. The first thing that makes this offer interesting is how founder-friendly the deal is, meaning that it is low dilution and is a huge token allocation for a small team.
>
> — [@anandPa94](https://x.com/anandPa94/status/2057618439867609295), May 21, 2026

The deal is the first time an AI lab has structured a batch-wide credit-for-equity instrument at this scale. AWS and Google Cloud have given YC companies $500k and $350k of cloud credits per batch in recent cohorts; both were marketing spend, not equity. OpenAI's $2M is four to six times larger and it asks for the cap-table line. For the field-check that opened this piece, the load-bearing fact is that every one of the 55 visible Spring 2026 companies, and the 116 unlaunched ones, now sits on the same equity instrument with the same counterparty. The Software for Agents lane (16 visible companies) and the Company Brain lane (4 visible companies) and the asymmetric Arlo slot are all on the same uncapped SAFE that converts at the next priced round.

That changes one specific part of the April 30 pre-mortem. The original piece argued that the most-funded category would have the highest mortality, with Software for Agents the obvious candidate, and the survival mechanism for the lane's 14-plus claimers was a vertical-wedge pivot or a mid-tier infra acquisition. The OpenAI deal changes the unit economics of that pivot window. A Software for Agents company that does not yet have product-market fit can now run agent inference against $2M of free GPT credits for the search-for-fit period. That extends the pivot window from "however long $500k of cash burns" to "however long $2M of GPT credits burns plus the cash runway." The structural effect on the lane is that the 18-month attrition prediction the April 30 piece made for SfA might still be right in *direction* but the timing is now elastic. A company that would have folded at month 12 on cash can run another six months on credits, which means the 2027 mortality data the April 30 piece said would settle the prediction is going to land in 2028.

The second-order effect is that uncapped SAFEs at the next priced round become the canonical cap-table structure for the entire cohort. OpenAI's eventual ownership floats up with whatever the Series A prices each company at. For the breakouts in the Top 12 above (Arlo, Hyper, Klaimee, Modern, Panacea, Astraea), that means OpenAI is the largest single non-YC stakeholder on the cap table at the first priced round, regardless of which RFS lane the company sits in. The OpenAI deal is, in this sense, the first market test that crosses every lane of the Spring 2026 batch simultaneously. The field check above is now also a tokenmaxxing experiment with 171 simultaneous trials.

{/* IMG-PROMPT: the-tokenmaxxing-variable: A single abstract company schematic on cream paper — a small box in the center labeled with a generic symbol, with a thick ink pipe flowing in from the left labeled (loosely) "$2M", and a thin red ink line flowing out from the right toward a smaller second box labeled (loosely) "equity". The cream paper carries faint grid lines like an editorial blueprint. Style: editorial illustration, cream paper background #f6f1e7, charcoal ink #1a1612, single editorial red accent #e63946, NO screenshots, NO text labels visible (or only one or two short ones), NO realistic logos, hand-drawn editorial quality. */}
![Tokens in, equity out. The same instrument now sits on every Spring 2026 cap table.](/post-images/2026-05-22-yc-spring-2026-rfs-three-week-field-check/the-tokenmaxxing-variable.jpg)

# The Sequoia $1T overlay

{/* IMG-PROMPT: sequoia-1t-overlay: An editorial illustration on cream paper showing TWO horizontal pools at very different sizes. The LEFT pool is small and labeled (loosely, hand-drawn) "$700B software"; the RIGHT pool is roughly 8-9 times larger and labeled (loosely, hand-drawn) "$6T services". A thin charcoal ink arrow flows from the small pool, arcing across the canvas, and lands inside the large pool. The arrow is the only editorial red accent. The composition reads as: SaaS spend bridging into the labor pool. Style: editorial illustration, cream paper background #f6f1e7, charcoal ink #1a1612, single editorial red accent #e63946, NO screenshots, NO realistic logos, hand-drawn editorial quality. */}
![The Sequoia ratio: $700B of global software spend versus $6T of professional-services spend. The arbitrage the autopilot thesis aims at.](/post-images/2026-05-22-yc-spring-2026-rfs-three-week-field-check/sequoia-1t-overlay.jpg)

The OpenAI offer reshapes the cap-table side of every Spring 2026 cohort company. A second, older variable reshapes the income-statement side. In April, Sequoia partner Julien Bek published [Services: The New Software](https://sequoiacap.com/article/services-the-new-software/), the essay The Deep Feed [decoded against the multiples math](https://www.thedeepfeed.ai/posts/2026-04-30-sequoia-services-as-software-thesis/) the day it came out. The single-line claim is *"the next $1T company will be a software company masquerading as a services firm."* The mechanism is a ratio. Global software is roughly $700B in spend. Global professional services is north of $6T. The SaaS era was built on the smaller pool; the autopilot era captures the larger one.

The two thesis frames talk to each other directly. The OpenAI deal answers the question of who pays for the inference. The Sequoia thesis answers the question of who pays for the work the inference does. The Spring 2026 batch is the first cohort where both answers are public when the company starts.

The Sequoia essay supplies a [founder checklist](https://www.thedeepfeed.ai/posts/2026-04-30-sequoia-services-as-software-thesis/#the-founder-checklist--does-my-company-fit-the-shape) of seven questions to test whether a pitch fits the services-as-software shape. The clean ones are: what does the customer's invoice say (per outcome or per seat), who signs the PO (an operating-line owner or an IT-budget owner), and which line of the customer's P&L you replace (labor or software subscriptions). A company that answers in the services-as-software column on five or more of the seven questions is the shape Bek is describing. A company that answers in the other column on three or more is, in his framing, a vertical SaaS company with AI features, which trades at one-tenth the multiple of the company it is pretending to be.

Apply that filter to the 16 RFS lanes and the lanes sort cleanly.

| RFS lane | Fits the services-as-software shape | Why |
| --- | --- | --- |
| AI-Native Service Companies | 🟢 By construction | Alströmer's lane is the explicit Bek lane. Pricing is per outcome, buyer is the line owner, P&L line replaced is labor. |
| AI Personalized Medicine | 🟢 Likely | Per-diagnosis pricing, per-trial pricing, regulated-license stack. The Astraea one-liner ("Agents that accelerate clinical trials") sits on the line. |
| Startups Selling to Huge Companies | 🟢 If the pitch is forward-deployed | The Hessian one-liner ("forward deploy with AI agents to automate business operations") is the Bek shape verbatim. |
| AI for Low-Pesticide Agriculture | 🟢 If priced per acre or per yield outcome | Garry Tan's lane is the cleanest possible outcome-billed surface, and zero visible Spring 2026 claimers. |
| Company Brain | 🟡 Only if sold to the manager who hires the knowledge worker | The four visible Spring 2026 claimers (Hyper, Ontora, Memory Store, Wato) all sell to the practitioner, not the manager. Pitched as the Sequoia shape, the lane is open. Pitched as a horizontal Q&A bot, it is the lane the April 30 piece said would lose. |
| Hardware Supply Chain | 🟡 Hybrid | Saudara AI's "AI sourcing broker linking brands to top global factories" reads as per-shipment outcome pricing, but the lane bleeds into BPO economics if the broker step requires a licensed human. |
| Counter-Swarm Defense | 🟡 Hybrid | Arlo Industries sells passive aerial sensing mesh, which is a hardware-led product, not an outcome-priced service. The fit depends on whether the broader Bosmeny vision (fused sensor and defender picture) is priced per-engagement or per-installation. |
| AI Operating System for Companies | 🟡 Almost certainly not | Hu's lane is a horizontal platform. Modern, ProjectX and Tasklet all sell tools to the company's operators rather than outcomes to the company's executives. |
| Dynamic Software Interfaces | 🔴 SaaS | Gigacatalyst's "customize your software to every user with AI" is a feature inside the existing seat-based pricing layer. |
| Software for Agents | 🔴 Infrastructure | The 16 SfA claimers sell tools to the people building agents. None of them is selling the agent's work to the buyer of the agent's work. The lane is the picks-and-shovels under the Bek thesis, not the thesis. |
| Inference Chips for Agent Workflows | 🔴 Infrastructure | Zero visible claimers, and if any apply, the lane is hardware infrastructure. |
| Supply Chain 2.0 for Semiconductors | 🔴 Mostly hardware / infrastructure | The lane is the substrate, not the autopilot. |
| Industrial Capabilities in Space | 🔴 Hardware | Space-industrial pitches do not bill per outcome at the office-line level. |
| Electronics in Space | 🔴 Hardware | Same. |
| SaaS Challengers | 🔴 By definition vertical SaaS | The lane is built on the assumption that the survivor is a better seat-based product. The Bek thesis explicitly buries vertical SaaS. |
| AI-Native Discovery Engines | 🔴 Mostly SaaS | Asterlab's "first AI-native AI research lab" reads as a tool, not an outcome. |

The split is striking. Roughly four of the sixteen lanes are clean services-as-software fits. Three are hybrids that could go either way depending on how the pitch is framed. Nine are not the Bek shape, and four of those nine are the saturated lanes (Software for Agents, AI Operating System, SaaS Challengers, Dynamic Software Interfaces) the Spring 2026 batch already crowded into.

That is the awkward read. The lanes that pulled the largest share of visible Spring 2026 claimants are the lanes that do not fit the multiples-math thesis Sequoia just put on the wall. The lanes that fit the multiples-math thesis are, with one exception (AI-Native Service Companies), the lanes the visible Spring 2026 batch under-claimed. Klaimee's *"liability insurance for AI Agents"* one-liner is the cleanest exception inside the saturated SfA lane, because insurance is an outcome-priced regulated business that happens to sit on top of agent infrastructure. The April 30 piece flagged Klaimee's wedge as the slot the lane was missing; the Sequoia overlay sharpens the read, because Klaimee is the only company in the SfA lane whose income statement should look like a services firm rather than a SaaS firm.

Panacea's *"AI-Native FDA Regulatory Services"* is the second exception. The one-liner reads as a Big 4 acquisition shape, not a vertical SaaS shape. The buyer is the pharma general counsel signing a per-filing engagement, not the regulatory affairs analyst buying a seat. The Sequoia checklist would score Panacea seven of seven, which is the rarest possible score in the visible batch and the company most likely to be the one the April 30 piece's pre-mortem flagged ("YC will fund one or two teams across the asymmetric items, and one of them will be the best company in the batch").

The third exception is Astraea, in Ankit Gupta's AI Personalized Medicine lane. The clinical-trials-agent one-liner is per-trial-outcome priced almost by construction, the buyer is the trial sponsor's medical-affairs office, and the line replaced is CRO labor. The Bek shape is intact. The April 30 read of the lane (medical founders run quiet, the strongest pitches apply privately) is consistent with Astraea being the visible front door and the rest of the lane being behind it.

The forward read for the Summer 2026 batch is that the next round of partner interviews is going to ask a question the Spring 2026 cohort largely did not have to answer. *"What does the customer's invoice from you say?"* A pitch that answers *"per closed book / per filing / per resolved interaction"* clears the Bek checklist's first question. A pitch that answers *"per seat per month"* does not, and a partner who has read Bek's essay knows the multiples math waiting on the other side of that answer. The OpenAI deal pays for the inference. The Sequoia thesis prices the work. A Summer applicant in the four clean services-as-software lanes (AI-Native Services, AI Personalized Medicine, Selling to Huge Companies, Low-Pesticide Agriculture) is walking into the intersection where both variables clear the bar.

# Twelve pitches that fit the May 25 deadline

The Spring 2026 batch left specific wedges empty inside the lanes the partners are openly recruiting. The twelve shapes below are the ones that survive both filters: a Spring 2026 visible-batch gap on one side, a Bek seven-question pass on the other. None of them is a company name. Each one is a wedge a Summer 2026 application can be built around between now and May 25. Where a Spring 2026 incumbent already sits in the adjacent slot, the incumbent is named so the applicant can pitch the complement rather than the alternative.

## In the four clean Bek-shape lanes

**One. SMB books-as-a-service for retiring-accountant towns.** AI-Native Service Companies. The customer is the SMB owner whose CPA retired and whose books have been on a single ledger for fifteen years. The invoice says *"per month-end close,"* the buyer signs as owner-operator, the line replaced is the fractional bookkeeper. Rillet and Basis sell into the CFO seat at growth-stage firms. The downmarket SMB shape, priced at $400-800 per closed month against a $4-8K accountant retainer, is the lane nobody in the visible Spring 2026 batch has claimed and the lane the 340,000-accountant shortage moves into first.

**Two. Per-claim adjudication for mid-market commercial insurance.** AI-Native Service Companies. EvenUp owns plaintiff-side personal-injury demand letters. The mirror surface, on the carrier side and in commercial mid-market, has no claimant in the visible Spring 2026 batch. The invoice says *"per resolved claim,"* the buyer is the head of claims at a regional carrier, the line replaced is the in-house adjuster's caseload. The regulated-license stack is the carrier's existing license, which the autopilot rides on.

**Three. Per-deal due-diligence for boutique private equity bolt-ons.** AI-Native Service Companies, with adjacent fit to Selling to Huge Companies. The customer is a lower-middle-market PE firm with twenty bolt-on acquisitions a year and a Big 4 contract it cannot afford to keep. The invoice says *"per closed transaction,"* the deliverable is the CIM extract, the QofE workpapers and the integration plan, the line replaced is the Big 4 transaction-services bill. The Spring 2026 batch did not claim it.

**Four. Per-read radiology second-opinion for community hospitals.** AI Personalized Medicine. Astraea sells per-trial outcomes to sponsors. The diagnostic-imaging-side surface, where community hospital radiologists send tumor-board cases out for second reads, is unclaimed. The invoice says *"per read,"* the buyer is the radiology department director, the line replaced is the locum radiologist bill. The regulatory pathway is FDA 510(k) for the read pipeline, which is a year and a half of work, which is exactly the asymmetric-silence shape Gupta said the lane has.

**Five. Per-patient-month titration for rare-disease specialty drugs.** AI Personalized Medicine. The customer is the specialty pharmacy or the rare-disease practice. The invoice says *"per active patient-month,"* the agent monitors side-effect profile and titrates dose against the prescribing rubric, the line replaced is the clinical pharmacist's case load. The n-of-1 frame Gupta named in the RFS is built on top of this shape, not the other way around.

**Six. Per-RFP-response for Fortune 500 sales orgs.** Startups Selling to Huge Companies. Hessian uses the *"forward deploy"* vocabulary the lane was written in. The specific service-as-software surface inside the lane is the RFP-response factory, billed per submitted RFP at a tenth of the price of the inside-sales engineer hours it replaces, sold into the CRO line. The Spring 2026 batch did not name it.

**Seven. Per-acre integrated-pest-management programme for specialty growers.** AI for Low-Pesticide Agriculture. The customer is a specialty grower (table grapes, almonds, leafy greens) whose pest-pressure cost line is one-fifth of their P&L and whose existing IPM advisor charges by the visit. The invoice says *"per acre per season"* for managed pressure under threshold, the line replaced is the agronomy retainer plus the chemical bill. The asymmetric-silence read on this lane is the most extreme in the entire RFS. Tan will fund whoever applies with an IPM contract on real fields.

## In the three hybrid lanes, done in the Bek shape

**Eight. Vertical Company Brain for pharma regulatory affairs.** Company Brain. The four visible horizontal claimers (Hyper, Ontora, Memory Store, Wato) sell to the practitioner. The vertical-RA wedge sells to the head of regulatory affairs at a mid-cap pharma. The invoice says *"per active programme,"* the line replaced is the contract-RA hours the team buys when a submission slips. The April 30 piece named the regulated-data ingestion wedge as the only survivor of the lane. The visible Spring 2026 batch left it empty.

**Nine. Per-incident response service for hospital IT security.** Company Brain, adjacent to AI Operating System for Companies. Modern sells the AI OS for IT broadly. The vertical-hospital-IT wedge inside that frame is incident response, billed per investigation, sold into the CISO line at a community hospital network. The line replaced is the MSSP retainer. Vertical-healthcare CISOs do not buy horizontal AI-OS pitches, which is the asymmetry.

**Ten. Per-engagement counter-drone targeting service.** Counter-Swarm Defense. Arlo Industries sells the passive sensing mesh. The complement is the intercept-decision service, billed per engagement under contract to a base operator, the line replaced is the operator-on-loop labor. The mesh is the input layer of Bosmeny's fused-picture RFS framing. The intercept service is the output layer the Spring 2026 batch did not claim.

## In the asymmetric silent lanes the Spring 2026 batch did not touch

**Eleven. Per-shipment export-compliance dossier for semicap tools.** Supply Chain 2.0 for Semiconductors. Zero claimants in the visible Spring 2026 batch, in a lane where US export controls on advanced fab equipment have made the compliance dossier the load-bearing step in every shipment. The customer is a semicap-tool OEM with twelve shipments a month to Asia and a Big 4 export-compliance contract that books two weeks per filing. The invoice says *"per shipment,"* the line replaced is the export-controls counsel and the BIS-licensing analyst time. Diana Hu's lane is silent for a reason and that reason is also the moat.

**Twelve. Per-mission integration test for cubesat manifests.** Industrial Capabilities in Space. Zero claimants in the visible Spring 2026 batch. The customer is a small-launch operator with a manifest of nine cubesats per launch and an integration-test cycle that takes three weeks. The invoice says *"per manifest,"* the deliverable is the completed integration-test pack and the regulatory-pathway dossier, the line replaced is the in-house integration engineer's six-week timeline. Oltean's lane is asymmetric for the same structural reason every other space-industrial lane is asymmetric: the founder population is in California machine shops, not on X.

# What this means for the Summer batch

The OpenAI offer closes May 25, which is the new YC Summer 2026 application deadline. The 48-hour deadline extension was announced by YC the day after the offer was published.

> OpenAI is offering $2M in tokens to every YC company in the spring and summer batches.
>
> We extended the summer deadline to May 25 so more founders can get in on it.
>
> — [@ycombinator](https://x.com/ycombinator/status/2057555656673210639), May 21, 2026

For a founder applying to Summer 2026 in the next 72 hours, the visible Spring 2026 evidence above is the most useful read on what the partner reviewing the application is going to be tired of and what they are going to be hungry for. The decision rules fall out of the field check directly.

Software for Agents is saturated. Sixteen of fifty-five visible Spring 2026 companies sit in the lane, which means a partner reading the Summer applications has already seen, internally, the same set of horizontal-framework, agent-auth, agent-sandbox and prompt-injection-defense pitches. A Summer applicant in the SfA lane needs to walk in with either (a) a primitive that is not one of the eight already-funded primitives, or (b) a vertical wedge that the Spring 2026 batch did not claim. The phrase "make something agents want" is, by the timing of Friedman's and Epstein's launch-week posts and the parody arc that followed, already a category the partner has discounted.

Counter-Swarm Defense has exactly one claimer in the visible Spring 2026 batch (Arlo). The asymmetric reading from April 30 still holds for the other four lanes (Industrial Capabilities in Space, Electronics in Space, Supply Chain 2.0, Low-Pesticide Agriculture), all of which have zero visible claimers. A Summer applicant in any of those five lanes faces the lightest possible competition for partner attention, and the partner authoring the lane (Bosmeny, Oltean, Johnston, Hu, Tan) is openly recruiting. The asymmetric upside reading from the April 30 piece is, in fact, the read that has held up best at three weeks.

The AI Operating System for Companies lane is contested but not saturated. Three Spring 2026 companies in the visible batch, one of them (Modern) with a clear vertical anchor. A Summer applicant pitching the AI-OS lane needs either a deeper vertical (legal, healthcare, finance) than Modern's IT vertical, or a primitive that the three already-funded teams do not own.

Vertical Company Brain, the wedge the April 30 piece said would be the only survivor of the lane, is unclaimed in the visible Spring 2026 batch. All four Company Brain claimers (Hyper, Ontora, Memory Store, Wato) are horizontal. A Summer applicant with a vertical Company Brain pitch (legal-data, healthcare-data, financial-data, regulated-content) is walking into a lane where the visible 32% of Spring 2026 left the vertical wedge empty. The two charitable readings of that gap (the vertical pitches are in the unlaunched 116, or the partners discounted Tratar's question) are both consistent with a Summer applicant being able to fill the slot.

AI-Native Service Companies is alive and saturated at the same time, which is the most awkward shape for a Summer applicant. Ten visible claimers, only three with a regulated-license stack in the one-liner. The April 30 pre-mortem read is the binding constraint: a Summer applicant pitching an AI-native service that does not own the license should expect the partner to ask, inside the first ten minutes, where the regulated-license-stack version of the pitch is.

The Summer batch will read the Spring batch the way the Spring batch read the RFS. Public claimers in the first cohort become the new chatter against which the second cohort's applications are filtered. The visible 55 above are now part of the launch-week noise the Summer applicant has to plan against.

# The next field check

The list was the prompt. The chatter was the answer key. The first cohort of claimers is now the field check on both.

Three weeks ago, the read on the Summer 2026 RFS was that launch-week founder engagement, OSS saturation, and partner amplification together painted a pre-mortem for the cohort. Eight of ten testable predictions held when measured against the 32% of the Spring 2026 batch that has launched public pages. The three that broke clustered around one structural mistake: under-estimating YC's tolerance for horizontal pitches in contested lanes. That single mistake explains the Company Brain break and the AI Operating System break in one sentence. The fifth asymmetric lane (Counter-Swarm Defense) produced exactly one company against a "zero" prediction; the one-or-zero budget still survived, but the strong-form wording did not.

The next field check is Demo Day. The 116 companies in the Spring 2026 batch that have not yet launched public pages will, over the next 30 to 60 days, do exactly that. Several of the broken predictions above have a charitable reading that depends on those 116 (the vertical Company Brain plays, the n-of-1 personalized-medicine plays, the inference-silicon plays) being heavier in the unlaunched cohort than in the visible 55. That hypothesis is testable on a known timeline. By the time the Spring 2026 Demo Day pages are public, the 32% sample becomes a 100% sample and every prediction in this piece becomes scoreable to the second decimal place.

Until then, the read holds at eight-and-three.

{/* IMG-PROMPT: the-next-field-check: A simple editorial calendar grid on cream paper — three columns of small ink boxes representing thirty days, with the final box at the bottom-right circled in a single editorial red ink line. Above the grid, a small ink clock face with hands pointing to the next hour. Style: editorial illustration, cream paper background #f6f1e7, charcoal ink #1a1612, single editorial red accent #e63946, NO screenshots, NO text labels visible (or only one or two short ones), NO realistic logos, hand-drawn editorial quality. */}
![A thirty-day calendar with the final day circled in red. The Demo Day field check is the next data point.](/post-images/2026-05-22-yc-spring-2026-rfs-three-week-field-check/the-next-field-check.jpg)

## Sources

- [The Deep Feed — YC's Summer 2026 RFS, read against the chatter (April 30 baseline)](https://www.thedeepfeed.ai/posts/2026-04-30-yc-rfs-summer-2026-decoded/)
- [The Deep Feed — Sam Altman writes a $2M token cheque to every YC company](https://www.thedeepfeed.ai/posts/2026-05-22-openai-2m-tokens-yc-batch-equity/)
- [Y Combinator — Requests for Startups](https://www.ycombinator.com/rfs)
- [Y Combinator — Spring 2026 batch listing](https://www.ycombinator.com/companies?batch=Spring%202026)
- [YC company page — Hyper](https://www.ycombinator.com/companies/hyper-4)
- [YC company page — Ontora](https://www.ycombinator.com/companies/ontora)
- [YC company page — Arlo Industries](https://www.ycombinator.com/companies/arlo-industries)
- [YC company page — Modern](https://www.ycombinator.com/companies/modern)
- [YC company page — Panacea](https://www.ycombinator.com/companies/panacea)
- [YC company page — Astraea](https://www.ycombinator.com/companies/astraea)
- [YC company page — Klaimee](https://www.ycombinator.com/companies/klaimee)
- [YC company page — Asterlab](https://www.ycombinator.com/companies/asterlab)
- [YC company page — Silmaril](https://www.ycombinator.com/companies/silmaril)
- [YC company page — Hessian](https://www.ycombinator.com/companies/hessian)
- [YC company page — Tasklet](https://www.ycombinator.com/companies/tasklet-2)
- [YC company page — Memory Store](https://www.ycombinator.com/companies/memory-store)
- [YC company page — Saudara AI](https://www.ycombinator.com/companies/saudara-ai)
- [@ycombinator — Spring/Summer batch token offer (May 21, 2026)](https://x.com/ycombinator/status/2057555656673210639)
- [@sama — tokenmaxxing startups (May 20, 2026)](https://x.com/sama/status/2056933166875857290)
- [@bosmeny — mic-drop moment (May 20, 2026)](https://x.com/bosmeny/status/2056914385814401238)
- [@TheRohanVarma — 2× Anthropic tokens framing (May 21, 2026)](https://x.com/TheRohanVarma/status/2057648423873270270)
- [@anandPa94 — Anand Paj, YC P26 founder, founder-friendly read](https://x.com/anandPa94/status/2057618439867609295)
- [TechCrunch — Sam Altman makes mic-drop offer to every Y Combinator startup](https://techcrunch.com/2026/05/20/sam-altman-makes-mic-drop-offer-to-every-y-combinator-startup/)
- [Business Insider — token offer is a new twist to startup investing](https://www.businessinsider.com/sam-altman-openai-offer-tokens-for-startup-equity-y-combinator-2026-5)
- [Hyperspell (YC F25) — Blomfield-anointed reference](https://www.ycombinator.com/companies/hyperspell)
- [Sequoia Capital — Services: The New Software (Julien Bek)](https://sequoiacap.com/article/services-the-new-software/)
- [The Deep Feed — Sequoia's services-as-software thesis, decoded](https://www.thedeepfeed.ai/posts/2026-04-30-sequoia-services-as-software-thesis/)

---

Canonical: https://www.thedeepfeed.ai/posts/2026-05-22-yc-spring-2026-rfs-three-week-field-check/
Site: https://www.thedeepfeed.ai
Full corpus: https://www.thedeepfeed.ai/llms-full.txt