The latest from Y Combinator

The latest from Y Combinator

Clueso raises $10 million to automate business video production

YC-backed Clueso announced a $10 million Series A led by Dharana Capital, bringing its total funding to $12 million. Founders Future and Neotribe Ventures joined existing investors Y Combinator and f7 Ventures; Clueso says more than 80% of the new capital came from existing backers.

More than 1,000 businesses use Clueso to turn workflows, documents, presentations and ideas into business videos. The company plans to invest in enterprise sales, AI agents, editing tools and integrations with company-knowledge systems, while expanding its staff from 35 to 50 by year-end.

Why it matters: Clueso is treating business video as an updateable workflow rather than a one-off generation task. Its next test will be whether refreshed content remains accurate and whether reported productivity and cost gains hold across its growing customer base.

The Economic TimesNewsbytesSuperpower Daily

AfterQuery reports coding gains from 500 expert-written tasks

YC-backed AfterQuery says 500 expert-written software tasks improved Alibaba’s Qwen3.8-27B-Medium on three coding benchmarks. The October 8 experiment fine-tuned the model with GRPO and compared it with the base model under the same testing setup.

The result reflects AfterQuery’s shift from finance agents toward expert datasets, evaluations and training environments for AI developers. However, all 500 tasks were used for training, with no held-out validation set, and the findings have not been independently replicated or tied to equivalent production gains.

Why it matters: AfterQuery is betting that specialized expert data—not another coding interface—is a scarce resource for improving workplace AI. Benchmark improvements are an initial signal; buyers will still need evidence that performance transfers to real codebases, reduces review time and meets security and cost requirements.

Streamline Feed

YC’s 2026 directory highlights AI for physical industries

Y Combinator’s October 13, 2026 directory pages show startups applying AI to operational work in sectors including real estate, construction, biotech and local services. The real-estate page lists 129 companies, while the biotech page lists 138; examples include HABU’s property-development operating system, Rasyn’s chemical-formulation predictions, Neuromorphic’s robots for existing wet labs and Cohesive’s sales and marketing automation for local services.

Why it matters: The directory points to a practical-AI pattern focused on specialized workflows rather than generic chatbots. These companies can be evaluated against concrete industry outcomes, but their claims still require validation through paying customers, repeat usage and reliable performance inside existing processes.

Y Combinator — Real estate and construction

BeSimple’s Pila8 offers paid voice tasks under lasting rights terms

BeSimple AI, a Spring 2025 YC startup, is recruiting contributors through Pila8 to record scripted reads, conversations and acted emotions for voice datasets. The platform displays rates before work begins and pays only for submissions that pass review; contributors can cash out through Stripe or Dots after Persona identity verification.

The arrangement gives BeSimple a perpetual, irrevocable license to recordings and voice likeness, including synthetic voices. The review describes typical pay of $25–$50 per hour, alongside reports of unclear rejections, daily caps and failed payouts; it also cites low app ratings and withdrawal reviews that can take up to five business days.

Why it matters: Pila8 provides a way to source specialized voice data while shifting approval, payout and long-term voice-use risks to contributors. For AI data operations, the key issues are transparent rights language, rejection explanations, task availability and payout reliability.

Aitraining

Advice for early-stage founders

General guidance: build the learning loop before the roadmap

Early-stage progress usually comes from reducing uncertainty, not maximizing feature count.

• Start with customer discovery. Talk to people who experience the problem regularly. Ask what they do today, what the workaround costs them, and what they have already tried. Look for repeated behavior and urgency—not just positive reactions to your idea.

• Build the smallest useful product. Create the narrowest version that helps a specific user complete a meaningful task. A manual step, limited workflow, or focused integration is acceptable if it lets you test whether the outcome matters.

• Prioritize by evidence. Rank work by its expected impact on a clearly defined user problem, the strength of the evidence behind it, and the cost of learning more. Defer requests that do not improve activation, retention, willingness to pay, or product reliability.

• Treat distribution as part of the product. Make it easy for the right users to discover, try, understand, and share the value. Choose channels where you can reach people with the problem, then measure whether attention turns into qualified conversations and repeat usage.

• Turn feedback into decisions. Separate requests from underlying needs. Track what users do, where they get stuck, and why they stop using the product. Review those signals regularly, make one clear change at a time when possible, and return to users to see whether the behavior improves.

The goal is a repeatable cycle: talk to users, ship a focused improvement, observe real usage, and update your priorities. Sustainability comes from learning faster than you expand the scope of the product.