ICYMI: Leopold, Space… | Sourcery Newsletter Summary

1. Key Themes

Autonomous AI Is Now Writing the Code, Not Just Assisting With It

The shift from AI as a coding copilot to AI as the primary engineer is no longer theoretical. Cognition reports that Devin writes roughly 95% of its own company's code, with total code shipped growing 7x in six months and customer usage up 11x–12x.

"Devin Writes 95% of Our Code" — Scott Wu, CEO of Cognition

"total code shipped has grown 7x in six months, and customer usage has increased 11x to 12x"

AI Infrastructure Investment Is Moving Beyond GPUs Into Optical Networking

The dominant narrative around AI infrastructure centers on Nvidia chips, but the article signals a material market shift toward the physical layer connecting those chips — specifically lasers and optics. Lumentum's stock rose ~1,000%, and the newsletter frames the opportunity bluntly:

"Everyone talks about Nvidia GPUs. Few people talk about the lasers connecting them."

"Copper is out, Optics are in"

Voice AI Is Scaling Into a Mass-Market Infrastructure Layer

Voice AI has moved from novelty to critical developer infrastructure. AssemblyAI now serves over one million developers processing approximately 100 million API calls per day, spanning AI notetakers, medical scribes, drive-through ordering, contact centers, and humanoid robots — signaling broad horizontal penetration.

"AssemblyAI went through Y Combinator's first AI batch in 2017 & now serves 1+ million developers with ~100 million API calls a day."

"1M+ Devs, 100x TAM, 4x YouTube"

Agentic AI Companies Are Raising and Scaling at Extraordinary Velocity

Cognition's financial trajectory illustrates the speed at which the agentic AI category is compressing traditional growth timelines. The company scaled annualized revenue from ~$37M to ~$492M in a single year while raising at a $26B valuation.

"growing annualized revenue from roughly $37 million to $492 million in a year... including a $1 billion round in May 2026 at a $26 billion valuation"

2. Contrarian Perspectives

The most critical AI infrastructure bet may not be compute — it may be photonics. The mainstream investment conversation is dominated by GPU supply chains and data center power. This newsletter argues the overlooked choke point is optical interconnects. Lumentum — a laser supplier, not a chip company — saw its stock rise approximately 1,000%, suggesting the market is beginning to reprice this layer of the stack.

"Everyone talks about Nvidia GPUs. Few people talk about the lasers connecting them." "Stock went up 1000%, Copper is out, Optics are in"

Autonomous software engineering may obsolete headcount scaling faster than expected. Conventional wisdom holds that AI accelerates engineering teams. Cognition's own internal data suggests something more radical: the AI is the engineering team. Writing 95% of code internally with a 7x output increase implies the marginal human engineer is rapidly depreciating in value at product companies that fully adopt agentic tools.

"Devin now writes roughly 95% of Cognition's own code, total code shipped has grown 7x in six months"

Leopold Aschenbrenner did not go to a16z — and that may matter for where AI safety talent flows. The newsletter drops a casual but notable aside: a widely circulated rumor that Aschenbrenner was joining a16z turned out to be false. His actual move — releasing an investor letter — signals he may be building independently rather than joining an established firm.

"Turns out, a16z didn't win this one. Damn. Nevertheless, congrats to Leopold on his wedding & investor letter."

3. Companies Identified

Cognition Applied AI lab; maker of Devin autonomous software engineer and Devin Desktop (formerly Windsurf) Why mentioned: Premier case study in agentic AI adoption and hypergrowth; eating its own cooking at 95% AI-written code

"The company has raised more than $2.5 billion, including a $1 billion round in May 2026 at a $26 billion valuation, while growing annualized revenue from roughly $37 million to $492 million in a year."

Lumentum (NASDAQ: LITE) Public company; supplier of optical networking and laser technologies for AI data centers Why mentioned: Positioned as the under-the-radar infrastructure play powering AI cluster interconnects as copper is displaced by optics

"Stock went up 1000%, Copper is out, Optics are in"

AssemblyAI Voice AI infrastructure platform; API-first speech-to-text and voice agent models Why mentioned: Scale benchmark for the voice AI infrastructure category; 1M+ developer reach across diverse verticals

"AssemblyAI went through Y Combinator's first AI batch in 2017 & now serves 1+ million developers with ~100 million API calls a day."

Brex Intelligent finance platform (cards, expenses, travel, bill pay, banking) Why mentioned: Newsletter sponsor; noted for serving high-growth AI companies including OpenAI and Anthropic

"Built for scale. Trusted by OpenAI, Anthropic, Vercel, Granola, Deepgram, & Sourcery.. teams that move fast AF."

MongoDB Developer data platform with operational, analytics, and AI-retrieval capabilities Why mentioned: Newsletter sponsor; cited for broad enterprise penetration

"~75% of the Fortune 100 – rely on MongoDB for their most important applications"

4. People Identified

Scott Wu CEO & Co-Founder, Cognition Why mentioned: Leading the most prominent agentic AI engineering company; spoke at RAISE AI Summit in Paris

"Devin Writes 95% of Our Code" — Scott Wu, CEO of Cognition

Michael Hurlston CEO, Lumentum (NASDAQ: LITE) Why mentioned: Executive voice on the optical infrastructure trend transforming AI data centers

"Michael Hurlston is the CEO of Lumentum (NASDAQ: LITE), one of the world's leading suppliers of optical networking and laser technologies powering AI data centers."

Dylan Fox Founder & CEO, AssemblyAI Why mentioned: Leading voice AI infrastructure platform from YC's first AI batch to 1M+ developer scale

"Dylan Fox is the founder and CEO of AssemblyAI, the voice AI infrastructure platform behind AI notetakers, medical scribes, drive through ordering, contact centers and humanoid robots."

Leopold Aschenbrenner AI researcher/investor (independent) Why mentioned: Widely watched figure in AI safety/AGI circles; rumored a16z hire that did not materialize; released an investor letter and got married

"Turns out, a16z didn't win this one. Damn. Nevertheless, congrats to Leopold on his wedding & investor letter."

5. Operating Insights

Use your own AI product as the primary engineering resource — and measure the output multiplier. Cognition doesn't just sell Devin; it runs on Devin. This "dogfooding at scale" strategy creates compounding advantages: real-world stress testing, authentic customer proof points, and a structural cost advantage over competitors still relying on traditional engineering ratios. The 7x output growth in six months is the KPI to benchmark against.

"Devin now writes roughly 95% of Cognition's own code, total code shipped has grown 7x in six months"

Build voice AI into products now — the developer infrastructure is mature and the TAM is repricing rapidly. With 100M API calls/day already flowing through AssemblyAI and use cases ranging from medical scribes to humanoid robots, voice AI has crossed the infrastructure threshold. The "100x TAM" framing suggests the market is still early relative to eventual size, meaning the adoption window for differentiated voice-native products remains open.

"Voice AI is Dominating... 1M+ Devs, 100x TAM, 4x YouTube"

6. Overlooked Insights

Leopold Aschenbrenner is entering the investment arena independently. The newsletter mentions his "investor letter" in passing, but the implication is significant: one of the most prominent AGI-timeline thinkers — who was dismissed from OpenAI and whose Situational Awareness essay shaped discourse across Silicon Valley — appears to be positioning himself as an independent investor or fund manager. Where his capital and intellectual influence land next is a high-signal event the newsletter treats as a footnote.

"congrats to Leopold on his wedding & investor letter"

AssemblyAI's customer list reveals where voice AI is already operating at scale in non-obvious verticals. The article briefly names humanoid robots as a use case alongside the more obvious notetakers and contact centers. This is a tell for where voice AI infrastructure is heading — into embodied AI — but it receives no elaboration.

"AssemblyAI, the voice AI infrastructure platform behind AI notetakers, medical scribes, drive through ordering, contact centers and humanoid robots"