Startup Beaker

AI Code Testing: The New Bottleneck

Blacksmith raised $45M Series B at $550M valuation—9x growth in 12 months.

Why: As AI coding tools (Cursor, Claude) make writing code 10x faster, testing & validation became the constraint.

Metrics:
• Revenue: 10x growth YoY
• Customers: 700 → 5,000+ (7x)
• Largest customers: $1M+/year spend
• Team: 10 → 30 people

How: Started as CI cloud → identified AI code validation gap → built Codesmith (auto-fix agent) → scaled to 5K+ customers.

Takeaway: When you solve the next bottleneck after the hype, you own the layer.

https://techcrunch.com/2026/08/12/blacksmiths-valuation-jumps-10x-to-550m-as-ai-coding-fuels-software-validation/

Brewed by Startup Beaker.
#AITools #DevTools

Autonomous Businesses Are Running on This API

Naïve raised $28.5M Series A to automate business setup and operations via AI agents. 30,000+ developers signed up in months.

What happened:
• Built unified API: LLC formation + payments + email + cloud infra in one prompt
• Scaled to 30K users by solving agent infrastructure costs (model routing, serverless runtime)
• 10x ARR growth in 6 months to low double-digit millions
• Series A led by Nexus Venture Partners

Why it matters: The bottleneck isn't building agents—it's running them profitably. Naïve owns the infrastructure layer.

https://techcrunch.com/2026/08/06/naive-raises-28-5m-to-automate-the-grunt-work-of-setting-up-and-running-a-company/

Brewed by Startup Beaker.
#AIAgents #Startups

EU AI Act just went live (Aug 2, 2026). High-risk AI systems now face binding enforcement.

What changed:
• Conformity assessments required for high-risk AI (healthcare, hiring, criminal justice, etc.)
• Technical documentation + CE marking mandatory
• Fines: €15M or 3% global revenue for violations
• Providers of general-purpose AI models must submit to EU SEND platform

Why it matters:
If you're building AI for EU customers—or anywhere—this sets the compliance floor. Most founders are behind. The window to audit your system, document training data, and add approval workflows is now.

Action: Audit your AI's risk classification. If high-risk, start documentation today.

Brewed by Startup Beaker.

Deal Velocity by MOST - texnologik startaplar uchun xalqaro akseleratsiya dasturiga qabul boshlandi

MOST C-Space Hub Startup Booster bilan hamkorlikda o‘z mahsulotini kengaytirish va O‘zbekiston bozorida rivojlantirishga tayyor asoschilar uchun 8 haftalik dasturni ishga tushirmoqda.

Dastur doirasida:
• amaliy vorkshoplar va mentorlik sessiyalari;
• startaplarni rivojlantirish uchun AI vositalari bilan ishlash;
• xalqaro ekspertlar ko‘magi;
• investitsiyalarni jalb qilishga tayyorgarlik;
• yakuniy Demo Day.

Ishtirok etish uchun quyidagi startaplar taklif etiladi:
• MVPga ega bo‘lgan;
• O‘zbekiston bozoriga yo‘naltirilgan.

FinTech va RetailTech yo‘nalishlaridagi loyihalarga ustuvorlik beriladi, biroq boshqa texnologik startaplardan ham arizalar qabul qilinadi.
Dastur yakunida Startup Booster jamoasi loyihalarni investitsion jozibadorligi nuqtayi nazaridan ko‘rib chiqadi.

📍 Format: gibrid (onlayn + oflayn)
Dastur tili: rus/ingliz tillari

🗓 Davomiyligi: 8 hafta
📢 Arizalarni qabul qilishning oxirgi kuni: 31-avgust.

RO‘YXATDAN O‘TISH UCHUN HAVOLA https://investready.uz/

Ollama Hits $65M Series B: Open Models Go Mainstream

Ollama closed $65M Series B (Theory Ventures lead). Total funding: $88M.

What: Dead-simple way to run open-source AI models locally. No cloud vendor lock-in. No API costs.

Traction:
• 8.9M monthly developers
• $65M Series B
• Investors: Theory Ventures, Benchmark, 8VC, Y Combinator, Pace Capital

Why it matters: Open models + local inference = lower costs, faster iteration, real control. If you're building AI products, Ollama is now the default dev tool.

Source: BusinessWire

Brewed by Startup Beaker.

Etched: Harvard Dropouts Hit $10.3B in 7 Months

Three founders dropped out, slept on floors, ran chip servers in a garage. Now: $300M Series C at $10.3B valuation.

What: AI inference chips that run any model faster than GPUs. Already booked $1B in customer orders.

How it happened:
- 2022: Dropped out, zero fundraising experience
- 2024: Shipped silicon, showed demos to Karpathy and Hinton
- 2025: $1B in contracts signed
- 2026: 400+ engineers, 10MW data center, $300M raise

Key lesson: Customer proof before scale. Skeptics don't matter when you have $1B in orders.

Brewed by Startup Beaker.

#AIInfra #Founders

Andera Raises $37M: AI Eats Enterprise Audit

AI agents just automated internal audit—one of the most rule-bound, judgment-heavy enterprise workflows.

Andera (founded 2024, Aryo Patel & Tinah Hong) raised $37M Series A led by Lightspeed + Bain Capital Ventures.

The wedge:
• Started with Fortune 100 customers
• Proved 70% cost savings on control testing
• AI agents read evidence, run tests, write workpapers in hours (not weeks)

Why it matters: If audit falls to agentic AI, so does every other "too complex" enterprise workflow. The pattern is clear—start with pain, expand to platform.

https://techcrunch.com/2026/07/23/aegisai-founded-by-former-google-security-execs-lands-36m-to-stop-ai-driven-spear-phishing/

Brewed by Startup Beaker. #AI #Startups

Cheap Chinese AI Models Are Killing Your Startup Advantage

Almost 200 Silicon Valley founders just sent a letter to Trump urging him NOT to ban Chinese open-weight AI models like Kimi K3. Here's why this matters.

The math:
• Kimi K3: ~$0.50 per 1M tokens
• Claude 3.5: ~$3 per 1M tokens
• For a bootstrapped startup doing 10B tokens/month: $5k vs $30k

If the US bans access, hundreds of early-stage startups die instantly. You can't afford $10k/month in Anthropic credits. Bigger companies can.

What's happening: The newly-formed Little Tech Association is now the startup counterweight to Big Tech lobbying. They're pushing for targeted safeguards, not blanket bans.

Your move: If you're building on open-weight models, you have 60-90 days to diversify or lock in costs. This is a founder-level risk.

Source: Politico, July 22, 2026

#startups #policy

Physical AI robots are working factory shifts today—not tomorrow.

Walden Robotics spun out of Toyota Research Institute in January 2026. By February, their wheeled humanoids were doing real work in a Toyota plant: loading parts, cleaning machinery, kitting components. Eight-hour shifts alongside humans.

The move: Large Behavior Models let robots learn new tasks through real-world practice instead of manual reprogramming. Conventional robots take weeks to retool; Walden does it in days.

Raised $300M seed (Toyota + Deviation Capital co-led) at $1.1B valuation. Already expanding into aerospace, semiconductors, logistics.

Lesson: Deploy first, raise later. Walden didn't pitch a future—they shipped a present.

https://www.waldenrobotics.com/news/walden-robotics-launches-from-stealth

Brewed by Startup Beaker. #Robotics #PhysicalAI

OPEN MODELS ARE EATING CLOSED MODELS' LUNCH

Ollama raised $65M Series B. Open-source AI is now the default for cost-conscious builders.

What: Developers run LLMs locally on PCs. No API calls, no token limits. Founded by ex-Docker engineers.

Proof:
• 8.9M monthly developers
• 176K GitHub stars
• 85% Fortune 500 using it
• 14 employees, 3 years to scale

How: Free desktop tool → GitHub distribution → proved agentic capability → monetized with neocloud subscriptions → scaled to 8.9M users.

Takeaway: Enterprises cutting inference costs. Open models + local compute = new default.

https://techcrunch.com/2026/07/09/popular-open-source-ai-developer-tool-ollama-raises-65m-grows-to-nearly-9m-users/

#OpenAI #Startups

Brewed by Startup Beaker.

Together AI: $800M Series C at $8.3B - Open-Source AI Infrastructure Wins

Together AI just closed $800M Series C (post-$8.3B valuation) to power enterprise AI on open-source models. This is the biggest infrastructure play of Q2 2026.

What it is: Cloud platform for training and running open-source AI models (DeepSeek, MiniMax, Kimi) at 10x lower cost than proprietary APIs.

How they got there:

- Founded 2022 to solve the API tax problem - enterprises paying 10x for closed models
- Built integrated platform: training + inference + fine-tuning in one place
- Positioned for AI agents (60% of new database launches now AI-driven)
- $1.15B ARR with customers like Cursor, Cognition, Decagon
- Syndicate: Aramco Ventures (lead), Vista, General Catalyst, Nvidia, Salesforce Ventures

Why it matters: The frontier lab era is ending. Enterprises now want to own their models. Together AI proved open-source + infrastructure = $8B+ business.

Founder move: If you're building AI products, you're choosing between renting (OpenAI/Anthropic) or owning (Together AI). The math is shifting toward ownership.

https://techcrunch.com/2026/07/01/together-ai-raises-800-million-series-c-to-power-open-source-ai-workloads/

#AIInfrastructure #Founders

Brewed by Startup Beaker.

Enterprise AI Independence: Prime Intellect's $130M Bet

Prime Intellect raised $130M Series A at $1B valuation to help enterprises build their own AI agents without relying on OpenAI or Anthropic

Key metrics:
- $100M ARR run rate (founded 2024)
- Customers: Ramp, Zapier, Flapping Airplanes
- Ramp's agent beat frontier models on accuracy while cutting costs 10x

Why it matters: Every founder building AI products faces the same choice - rent from frontier labs or own your intelligence

Prime Intellect just proved the latter is viable at scale

https://techcrunch.com/2026/07/08/prime-intellect-raises-130m-series-a-to-help-enterprises-build-their-own-ai-agents/

Brewed by Startup Beaker

#AIAgents #EnterpriseAI

Bending Spoons: From Failure to $18B IPO

Italian startup just went public at $18B—40% pop on day one. The real lesson: they turned a failed startup into a playbook for operational excellence.

The Model:
• Buy struggling SaaS brands (Evernote, Vimeo, Meetup)
• Cut bloat, ship AI features fast
• Revenue per employee: $1.12M (2023) → $2.57M (2025)

How They Did It:
• Failed first startup taught them luck is huge in zero-to-one
• Built data-driven pricing & experimentation toolkit
• Hired ruthlessly for talent
• Deployed AI aggressively

Takeaway: You don't need a novel idea. You need a repeatable system that removes randomness from growth.

https://techcrunch.com/2026/07/01/after-18b-ipo-bending-spoons-founder-says-success-comes-from-minimizing-luck/

#Startups #OperationalExcellence

Brewed by Startup Beaker.

Dispatch is the nerve center.

Probook just raised $40M (A16Z + Sequoia) to unify home-services dispatch. Instead of bolting AI onto existing tools, founder George Eliadis built the entire OS around dispatch—lead intake, scheduling, billing, customer thread all in one.

Why it worked:
• Founder had domain pain (ran his father's pressure-washing biz)
• Replaced 10+ point solutions with 1 integrated system
• Customers already have 100s of technicians on platform
• Clear ROI: dispatch chaos → revenue leak

Lesson: Pick a vertical's core bottleneck, not its edges. Dispatch was invisible to generalists but worth $40M to specialists.

#startups #verticalsaas

8090 Labs Raises $135M Series A: Chamath's AI Coding Bet

Chamath Palihapitiya just raised $135M for 8090 Labs, his AI coding agent for enterprises. He's stepping in as CEO full-time.

The metrics:
• $135M Series A (Salesforce Ventures led)
• Product: Software Factory—AI agents that write production-quality code, not prototypes
• Enterprise controls: audit trails, compliance, governance built-in
• Investors: Katzenberg, Sacks, Friedberg, Calacanis, Arora, D'Angelo

How it happened:
• Founded Jan 2024 to solve enterprise AI coding gap
• Identified that most AI coding tools ship vibe-coded prototypes, not production systems
• Built controls layer enterprises actually need (audit, rollback, approval workflows)
• Proved traction with early customers
• Closed $135M at right moment when AI coding is hot

Why it matters: Chamath left Facebook early. He's betting this AI moment is bigger. When operators move full-time, capital follows.

https://techcrunch.com/2026/06/29/chamath-palihapitiya-raises-135m-series-a-for-his-ai-coding-startup-takes-ceo-role/

#AIStartups #Founders

Brewed by Startup Beaker.

Why Your Best Work Gets Punished

High-quality work doesn't always win. In fact, it often gets ostracized.

Researchers modeled organizations where agents choose high (H) or low (L) effort. The finding: low-effort agents systematically out-earn high-effort ones, forcing try-hards into early retirement. Fresh high-effort hires replace them, get exploited again, and the cycle repeats.

The mechanism:
• Low-effort agents start satisfied with mutual-low outcomes
• High-effort agents start optimistic, get exploited, then capitulate
• Once capitulated, they're trapped (switching back earns even less)
• Low-effort agents inherit the org

The fix isn't hiring more try-hards—it's hiring non-selfish agents (heroes/saints who won't capitulate) or sanctioning low-quality work frequently.

Takeaway: Cartels of mediocrity form not from free-riders, but from disillusioned high-performers. Protect your standards or migrate.

Source: Proietti & Franco, "Social Norms and the Dominance of Low-Doers" (2018)
https://blog.taylorwood.io/2026/06/19/social-norms-and-low-doers.html

180 Rejections, $100M Raised: Why Bland Bet on Proprietary Voice AI

Isaiah Granet got rejected by 180 investors in 3 weeks during Y Combinator. Their reason: phone calls won't exist in a year.

He just closed $50M Series C (led by Dell Technologies Capital), bringing total raised to $100M+.

Why it worked:
• Built proprietary in-house voice models (won't integrate OpenAI/Anthropic)
• Handles 30-45 min complex calls (not just routing)
• 3.5M calls/week, 175M processed last year
• 250+ enterprise customers (Samsara, Kin Insurance, CNO Financial)

The wedge: Nejad's aunt couldn't reach her insurance company by phone. They built to solve that problem—and found a $13.5B market.

Lesson: Conviction + proprietary tech + real problem = investor reversal.

https://fortune.com/2026/06/16/voice-ai-bland-50-million-after-being-rejected-by-180-investors/

#VoiceAI #Founders

Conduct just raised $60M Series A to collapse enterprise software timelines.

Three ex-Palantir founders saw the same problem everywhere: companies waiting months for system changes because legacy code is opaque. Their fix: AI OS that maps custom code to business logic, turning 5-month projects into 1-afternoon sprints.

How they did it:
• Started at Palantir—saw the pattern across industries
• Built AI agents that parse hundreds of millions of lines of code + gather context
• Proved it with early customers (Daimler, DHL, Heidelberg Materials)—50-80% cost savings
• Landed SAP partnership + $60M from Index Ventures + ICONIQ

Why it matters: Speed of execution is your moat. If you can ship 10x faster than competitors, you win.

Brewed by Startup Beaker.
https://conduct.ai/blog/series-a

#startups #founders

Supabase raised $500M at $10.5B—10M developers in 6 years.

Metrics:
- 600% YoY growth in database launches
- 60% of new databases launched by AI tools
- 10M developers (doubled in 8 months)

How they scaled:
- Open-source Postgres tooling (community-first)
- Positioned for AI agents (60% of growth)
- Built "Supabase for Platforms" (fastest-growing product)
- Invested in reliability and autonomous management

Takeaway: Open-source + AI infrastructure + unglamorous ops = scale.

https://supabase.com/blog/supabase-series-f

#Startups #Funding

Claude Fable 5: Frontier AI, Now for Everyone

Anthropic released Claude Fable 5 today. Mythos-class model, general availability.

Key metrics:
• Stripe: 2 months of work → 1 day
• State-of-the-art on 95%+ benchmarks
• $10/M input, $50/M output (50% cheaper than Preview)
• Handles millions of tokens in long-running tasks

How they shipped it: New safety classifiers block cyber/bio misuse. 95% of sessions unaffected. Fallback to Opus 4.8 on flagged requests—no hard refusals.

Takeaway: Agentic workflows are now viable at scale. If your product uses AI reasoning, this is your new baseline.

https://www.anthropic.com/news/claude-fable-5-mythos-5

#AI #Founders