Retell AI reports $80 million ARR; YC lists 136 robotics startups; Gumloop builds enterprise-agent sales
Today’s developments show AI startups moving from product proof toward repeatable adoption: Retell AI and Gumloop are strengthening enterprise sales, while Firecrawl formalizes customer engagement and Parallel Bio expands validation work. The pipeline is also widening, from YC’s robotics and crypto companies to Memorable’s procedural memory for agents, putting more emphasis on reusable workflows, measurable performance, and deployment.
Saturday, Sep 19, 2026
Tracking: Y Combinator
Geography: United States, San Francisco Bay Area
YC-backed Retell AI says revenue reached $80 million as hiring expands
Y Combinator-backed Retell AI says its annual recurring revenue has grown from $5 million at the start of 2025 to $80 million.
The Redwood City voice-AI company also says it is valued above $1.5 billion and is hiring an enterprise account manager to manage renewals, customer expansion, and usage growth.
The role signals a shift from proving the product to systematically increasing enterprise adoption.
A separate YC listing shows a different operating model: Educato, a YC S24 education startup, is hiring a full-time product-and-AI engineer in Bucharest to work directly with its founders.
Together, the listings show YC companies at sharply different stages, from founder-led product building to enterprise monetization, while both continue investing in AI applications.
Key facts:
Retell AI says annual recurring revenue rose from $5 million to $80 million since early 2025.
The Redwood City company says it is valued above $1.5 billion.
Retell is hiring an enterprise account manager focused on renewals, expansion, and usage growth.
Educato, a YC S24 company, is hiring a full-time AI-focused engineer in Bucharest.
Why it matters: Retell’s advertised numbers are company claims in a recruiting listing, but they describe a meaningful change in startup priorities: enterprise retention and expansion now sit alongside product development.
Its usage-based pricing model makes customer consumption central to revenue growth, so the new hire is tasked with turning pilots into broader deployments and renewals.
Educato’s listing provides a contrasting YC pattern: direct founder access, broad engineering ownership, and fast product iteration at an earlier-stage company.
The next point to watch is whether Retell can convert reported growth into durable enterprise retention, while smaller YC startups continue using lean teams to build AI products.
YC’s September directory highlights new robotics and crypto startups
Y Combinator’s September 19 directory pages offer a current snapshot of its startup pipeline in two frontier categories: robotics and cryptocurrency.
The robotics page covers 136 companies, while the cryptocurrency page highlights eight; both include firms from recent YC cohorts, including F2026 and S2026.
The featured robotics companies span general-purpose machines, cloud inference, scientific experimentation, defense exoskeletons, and rapidly deployable power systems, with many based in San Francisco.
Crypto entries focus on AI-assisted investing, stablecoins, payments, and on-chain data.
One robotics company, Edgerun, says it has built more than a dozen exoskeleton iterations since April, tested them with Ukrainian fighters, and secured more than $150 million in letters of intent.
Key facts:
YC’s September 19 robotics directory page covered 136 companies.
Its cryptocurrency page highlighted eight YC-funded startups.
Five featured F2026 robotics startups are based in San Francisco.
Edgerun says it secured more than $150 million in letters of intent.
Instinct is a five-person, San Francisco-based W2026 crypto startup.
Why it matters: The snapshot points to broad early-stage activity around physical AI and crypto financial infrastructure.
Robotics entries connect AI to factories, data centers, science, defense, and energy, while crypto companies target trading, stablecoins, payments, and on-chain data.
For investors and founders, the highlighted teams’ small size makes early validation and financing central tests.
Edgerun’s letters of intent are an unusually concrete commercial signal in the sample; the next question is which other companies convert ambitious descriptions into deployments, customers, or financing.
Firecrawl posts San Francisco events role as AI infrastructure startup scales
Firecrawl, an AI web-data company, posted a San Francisco events-producer role on September 19, offering $155,000 to $174,000 annually plus competitive equity. The position is on-site five days a week, includes travel, and requires U.S.
work authorization without current visa sponsorship. The hire would own a monthly community series at Firecrawl’s San Francisco headquarters, private dinners for customers and prospects, conference activations, and team offsites.
The role also covers venues, vendors, budgets, logistics, merchandise, and post-event attendance and cost tracking.
The company says it reached eight-figure annual recurring revenue in its first year, more than doubled revenue in its second, and has more than 180,000 GitHub stars.
Those figures present Firecrawl as a fast-growing AI infrastructure provider seeking to institutionalize how it engages developers, customers, partners, and employees.
Key facts:
The listing was dated September 19, 2026.
Firecrawl offers $155,000–$174,000 annually plus competitive equity.
The role is based at Firecrawl’s San Francisco headquarters, onsite five days weekly.
The producer will run monthly community events, private dinners, conferences, and team offsites.
Firecrawl says it exceeded eight-figure ARR in year one and more than doubled in year two.
Why it matters: This is a hiring signal, not evidence of a new financing, acquisition, or YC batch announcement.
Firecrawl is putting dedicated capacity behind recurring community programming and targeted gatherings while requiring the operator to measure attendance, cost per attendee, and follow-through—an indication that events are being treated as a repeatable growth and relationship channel, not one-off marketing.
Founders and investors should treat the revenue and GitHub figures as company-provided operating metrics, not independently verified results.
The role’s fully onsite structure and lack of current visa sponsorship also limit the candidate pool as Firecrawl expands its external presence.
Gumloop cites more than $70 million raised as it builds enterprise AI-agent sales
Gumloop, an AI automation startup backed by Y Combinator, is hiring a pre-sales solutions engineer while presenting itself as an enterprise platform for deploying AI agents.
In a September 19 job posting, the company said it has raised more than $70 million from Benchmark, First Round Capital, Y Combinator, Nexus Ventures, Shopify Ventures, BoxGroup and The Cannon Project.
It also said teams at Gusto, Ramp, Shopify, Samsara, Instacart and Opendoor are already using the platform to automate work across their organizations.
The role is designed to help account executives win deals by running technical discovery, building tailored demos and proofs of concept, and piloting customers through production.
Gumloop describes its product as a combination of conversational agents, visual workflows and governance tools, including managed MCP hosting.
The hiring effort points to a company trying to turn reported customer deployments into repeatable enterprise sales while giving customers control over their own automations.
Key facts:
Gumloop’s September 19 posting says the company has raised more than $70 million.
Investors named include Benchmark, First Round Capital, Y Combinator and Shopify Ventures.
Gumloop says Gusto, Ramp, Shopify, Samsara, Instacart and Opendoor run agents in production.
The open role owns technical discovery, demos, pilots and customer deployment.
Why it matters: Gumloop’s posting captures a broader test facing enterprise AI startups: converting impressive agent demonstrations into dependable, organization-wide deployments.
Its sales-engineering role is built around security questions, integrations, proofs of concept and customer training, suggesting that adoption depends on implementation as much as on model performance.
For Gumloop and its investors, named production customers provide a basis for expanding sales.
The next important signals will be whether those deployments translate into additional contracts and whether the company can scale customer implementation without weakening its governance and security controls.
YC S27 startup Memorable launches procedural memory for AI agents
Memorable, a Y Combinator S27 startup, has launched a procedural-memory system for AI agents, according to a September 19 LinkedIn post.
Founded by Advaiyt Sane, a CMU computer science graduate, and Nikhil Krishnaswamy, a Stanford computer science graduate, the company says its software turns successful agent traces into a knowledge graph of reusable procedures.
It then retrieves and assembles relevant steps when similar work returns, rather than forcing an agent to start from zero. Memorable says the system can also infer workflows no agent has completed end-to-end by combining prior tool calls.
Its CLI is live across Claude Code, Codex, Cursor, and Claude Cowork.
The launch frames the product as a layer that helps companies retain operational experience—how agents are debugged, failures repaired, and successful paths reused—alongside their models and data.
Its stated goal is to make agents faster, cheaper, and more deterministic by treating reasoning as graph search.
Key facts:
Memorable is identified as a Y Combinator S27 startup.
Founders Advaiyt Sane and Nikhil Krishnaswamy created the company.
Its CLI is live across Claude Code, Codex, Cursor, and Claude Cowork.
Memorable turns successful agent traces into a reusable knowledge graph.
The launch was highlighted in a September 19 LinkedIn post.
Why it matters: Most current agent work, as Memorable frames it, begins inside one context window and may be repeated from scratch.
Its approach shifts attention from models and data toward accumulated operational experience: procedures, debugging paths, and repairs that can be reused.
If the system performs as described, teams could pursue more consistent and less repetitive agent workflows without retraining the underlying model.
The launch offers a concrete example of an early-stage startup building infrastructure around agent reliability rather than another standalone model.
The next practical question is whether procedures learned from earlier work can be reused safely when an agent encounters unfamiliar tasks.
YC-backed Parallel Bio recruits for organoid drug-discovery buildout
Parallel Bio, a Y Combinator-backed drug-discovery startup, posted four roles on September 19 for its Vienna and San Francisco operations.
The openings span immunology, iPSC and organoid research, bioengineering, and a San Francisco-based platform-science leadership post, showing how the company is moving its research platform toward repeatable external work.
The company says it has raised about $30 million from AIX Ventures, Marc Benioff, Jeff Dean, and Y Combinator, and that pharma partners, including three Fortune 500 companies, use its platform.
Parallel Bio says its immune-organoid and AI system is 87% concordant with clinical patient data; the hiring emphasis is now on validating models, improving reproducibility, designing multi-organoid formats, and transferring assays into routine execution.
Key facts:
Parallel Bio posted four Y Combinator roles on September 19, spanning Vienna and San Francisco.
The company says it raised about $30 million from AIX Ventures, Benioff, Dean, and Y Combinator.
Parallel Bio says its platform is 87% concordant with clinical patient data.
Three Fortune 500 companies are among pharma partners using the platform.
The San Francisco role leads assay development, technology transfer, controls, and routine external execution.
Why it matters: The hiring pattern points to a startup shifting from proving its biological models to making them dependable at scale.
Vienna roles focus on building immune organoids, co-culture systems, assays, and custom hardware, while the San Francisco leadership role is tasked with reference data, acceptance criteria, and transferring assays into routine partner work.
For investors and prospective customers, the next test is operational rather than purely conceptual: whether Parallel Bio can preserve complex human biology while making experiments reproducible across scientists, sites, and automated workflows.
Its stated pharma relationships and regulatory tailwinds provide momentum, but the company’s own roles show that standardization and validation remain central execution challenges.