Gutenberg Lab — Morning Paper

AI Coding Agents Speed Development but Don't Expand What Software Can Do—TypeSafe AI's JEV Offers a New Programming Primitive

Featuring

Martin Casado (General Partner, a16z) × Ben Horowitz (Co-founder and General Partner, a16z) × Diogo Almeida (Founder and CEO, TypeSafe AI)

Bottom line

AI coding agents write existing code faster, but they don't expand what software can actually do.

Three-Line Summary

TypeSafe AI founder Diogo Almeida criticizes coding agents like Claude and Codex as "just-in-time software" that merely accelerate the production of the same code humans would write. His product, JEV, is a new programming primitive that accepts natural language intent and returns probabilistic decisions, embedding intelligence inside software itself. OpenAI has been trying to automate customer service since 2020 without success, and Almeida argues the AI industry has optimized for impressive demos to humans rather than actual automation.

Three Key Points

1. Coding agents are good at syntax but terrible at architecture: Almeida states that "coding agents are really good at syntax but incredibly bad at architecture," which is "the most human creative part of software." He estimates agents are at the "fiftieth percentile at architecture," suggesting a trade-off between speed and quality in system design.

2. Reliability is the entire product: TypeSafe delayed launch for years to fight for "weird nines of reliability," believing "every nine of reliability is going to be so valuable" in enabling new applications. Almeida defines four layers of reliability, with the ultimate goal being developers trusting JEV enough to program against it without making example queries first.

3. The SaaSpocalypse narrative was backwards: While coding agents caused SaaS valuations to drop on fears that software would become trivially replicable, Almeida predicts "SaaS will be one of the largest winners" of AI. SaaS companies already have customer relationships and deep workflow knowledge, and their capital investment is in customer acquisition—if they make their software way better with JEV, that's a huge advantage.

Editorial Perspective

The observation that coding agents increase developer productivity without expanding software capabilities cuts to the heart of platform transitions. During the shift from the birth and rise of the internet to the App Store economy, development tools always evolved, but true transformation came from new primitives—touch interfaces, location services, cameras. JEV's proposed primitive of "accept natural language, return probabilistic decisions" could expand the boundary of what software can execute. However, as Almeida notes in delaying launch "for years" to prioritize reliability, the value of a new primitive is determined by implementation trustworthiness. Not demo flashiness, but the reliability to run unsupervised will decide whether this transition succeeds.

Source: a16z, "AI Can Write Code. Why Isn't Software Better?" (September 28, 2026)

https://a16z.simplecast.com/episodes/ai-can-write-code-why-isnt-software-better-je38wEkC

Spoken source: a16z, "AI Can Write Code. Why Isn't Software Better?" (September 28, 2026). Guest: Diogo Almeida.

AI Disclosure: Produced with AI assistance; facts and analysis reviewed by our editorial team.

SpaceX's Louisiana Starbase: Sun-Synchronous Orbit as Interplanetary Staging Point

Featuring: Anthony Colangelo (Host, Main Engine Cut Off)

Bottom line: SpaceX's Louisiana site will enable high-cadence launches to sun-synchronous orbit for data center satellites, with leftover propellant accumulating in co-planar depots that serve as staging points for lunar and Mars missions via three-burn departures.

3-Line Summary

• SpaceX announces $100 billion Louisiana launch site with direct sun-synchronous orbit access, critical for data center satellite constellation and impossible from Texas
• Falcon 9 commercial launches end in 2028 as SpaceX shifts entirely to internal payloads (Starlink, data centers), creating rare market opening for next-gen launchers
• Stoke, Rocket Lab Neutron, and Relativity all approaching flight readiness in 2026-2028 window, precisely when SpaceX withdraws from commercial launch services

3 Key Points

1. Trajectory advantage: Louisiana offers direct sun-synchronous orbit trajectory (1,000-1,200 km over water before Mexico landfall), polar orbits, and multiple mid-inclination corridors. Cape Canaveral requires "dogleg maneuver" around Cuba with performance penalty; Texas can only fly north or south of Cuba. Louisiana is "shockingly good" and the first SpaceX site without major qualifiers

2. Orbital refueling architecture: Drawing on research by Michael Lauks and John Goff, Colangelo proposes co-locating propellant depots with high-frequency flight orbits. Leftover propellant from data center satellite launches accumulates in depots at different altitudes but same inclination, enabling lunar/Mars missions without dedicated tanker flights. With refueling, performance losses from non-optimal departure inclinations become "negligible"

3. Landing safety at scale: Sun-synchronous orbit from Louisiana creates northbound reentry corridor entirely over Gulf of Mexico after 12-hour orbital mechanics. At SpaceX's target scale of "thousands per year," this offers safer and more predictable landing opportunities than mid-inclination orbits. Even lowest-inclination trajectory has ~500 km water buffer

Editorial Perspective

SpaceX must become "the biggest company in the world in multiple sectors" to achieve their launch cadence goals — petroleum/natural gas production, shipping logistics, satellite manufacturing, data centers. Louisiana's access to the cheapest natural gas pipelines in the US solves the absurd logistics of Starbase Texas, where "15,000 trucks" are needed to fuel a Starship. This $100 billion industrial buildout is SpaceX replicating their aerospace outlier status in multiple other industries.

Source: Main Engine Cut Off "T+339: Starbase Louisiana, the Falcon Fade, and the Next Wave of Launch Vehicles" (August 28, 2026)

https://mainenginecutoff.com/podcast/339

AI Disclosure: Produced with AI assistance; facts and analysis reviewed by our editorial team.

• --

Bitcoin at $85,000 — A Bear's Capitulation Reveals the Shallowest Downturn Ever, and the Limits of a Bull Market Without Liquidity

Featuring

David Hoffman (Bankless co-host) × Ben Cowen (crypto analyst, Into the Cryptoverse)

Bottom line

This bear market was the shortest and shallowest on record, which means the next bull market will likely deliver "diminishing returns" rather than the explosive rallies of past cycles.

Three-Line Summary

Crypto analyst Ben Cowen admitted he was wrong about Bitcoin failing to break its May high. This bear market lasted only nine months and spent just 100 days below the 200-week moving average (versus 270 days last cycle). The reason: "fewer sins to pay for" — no Terra Luna or FTX-scale collapses, and the most favorable regulatory environment ever. However, global net liquidity remains 17% below its 2021 peak at $25 trillion, and Alt/Bitcoin pairs continue their five-year downtrend. Cowen argues that a true "alt season" requires a crisis to force money printing, but such a crisis would hurt crypto first — the catch-22 of this cycle.

Three Key Points

1. Bitcoin likely bottomed with its smallest-ever 53% drawdown. Last cycle saw 270 days 30% below the 200-week moving average; this cycle saw only 100 days at single-digit percentages below. The difference: no catastrophic collapses (Terra Luna, Three Arrows Capital, FTX) and "the most favorable regulatory conditions of all time" instead of Gary Gensler's assault. But a shallow bottom implies a "more modest bull market" — don't expect the 7-8x rallies of last cycle.

2. Accommodative monetary policy ≠ loose liquidity — the former is lifting altcoins now, the latter's absence has suppressed them for five years. The Fed funds rate at 4% is 90 basis points below the neutral rate (approximated by the 2-year yield at 4.9%), making policy accommodative for the first time since 2022. This explains recent altcoin strength. But global net liquidity (sum of major central bank balance sheets) is ~$25 trillion, down from ~$30 trillion in 2021 — a 17% decline. A 2020-2021 style "alt season" requires a liquidity surge, which requires a crisis.

3. The YouTube view surge is an illusion — a metric change, not returning retail interest. In mid-August 2026, YouTube changed its view-counting threshold from ~30 seconds to one second (matching X, Instagram, TikTok). Cowen's views doubled overnight, but his actual audience didn't grow. Other metrics (YouTube subscribers, Twitter/X followers, Coinbase app rankings, Google Trends, Wikipedia views) remain flat, showing retail hasn't returned.

Editorial Perspective

The four-year cycle isn't dead, but it's morphing. A shallow bottom means a lower, earlier top — a sign of maturation that will disappoint those expecting a 2021 replay, but reward those who adjust expectations. Bull markets without liquidity are possible, but they don't explode. The market still moves forward, just differently.

Source: Bankless "Ben Cowen Says You Have Permission to be Bullish" (September 28, 2026)

http://podcast.banklesshq.com/

AI Disclosure: Produced with AI assistance; facts and analysis reviewed by our editorial team.