AI Coding Assistants

Today's Stories

1. 1

   Claude Code's Boris Cherny: Black-box AI code OK only for prototypes

   Claude Code's creator Boris Cherny answered a developer's question on September 11, 2026, saying throwaway prototype code can be treated as a black box, but production code should be held to a higher bar than human-written code. It draws a practical line between prototype and production use, with Anthropic itself relying on Lint, E2E tests, fuzzers and code review — read as a quality bar developers are expected to hold, not lower.

   The approach hinges on tests and specs being trustworthy, since tests like expect(true).toBe(true) can pass without being meaningful; the writer suggests starting small with reliable tests and the latest specs.

2. 2

   ITR analyst: Japan firms 'almost absent' from harness engineering

   ITR principal analyst Hiroaki Koumoto said Japanese firms' efforts in harness engineering are 'almost nonexistent.' Only some leading IT vendors and startups use AI agents in software development, he said, while typical companies and SIers are still stuck at an earlier stage. Without an integrated agent development process, building a harness has little meaning, so many Japanese firms may keep treating AI as a tool for cutting coding hours rather than a path to better competitiveness.

   Handling AI agents safely and securely company-wide, not individual tools like MCP, is the definition Koumoto uses — and he sees a roughly five-year window before the gap with global firms becomes hard to close.

3. 3

   Momentic launches Mo, an AI agent that tests apps without scripts

   Momentic Inc. launched Mo, an AI agent that opens an app from a URL and prompt, spins up a swarm of agents to try thousands of permutations and edge cases, and reports failures with video evidence. The launch removes the need for developers to write and maintain test scripts, which the company says become a liability every time an app changes, so QA teams can work from requirements and intent directly.

   Whether Mo can replace scripted suites in practice hinges on how well its agent swarm handles complex apps, where the article says more tests are needed and eventually become impossible to keep up. Watch customer results from Notion, Superpower, Iris, Committee for Children and Boundless, which have been putting Mo through its paces over the past few weeks.

4. 4

   Meta launches Meta Enterprise Platform to sell AI

   Meta is launching the Meta Enterprise Platform, a new business unit selling the Muse agent, Meta Business Agent, Muse API and Muse Code to companies, led by former MongoDB CEO Chirantan Desai, reporting to Mark Zuckerberg. Meta is moving into the enterprise AI market against coding tools like Claude Code, Codex and Cursor, and cheap Chinese open-weight models, so its AI could become a paid business rather than just an internal investment.

   Meta hasn't said yet exactly how the business will work or what the services will cost, so the plan hinges on pricing and packaging. Watch whether the unit can turn Zuckerberg's claimed edge in models, agents and infrastructure into paying customers.

5. 5

   Atlassian CEO Mike Cannon-Brookes: the SaaSpocalypse that wasn't

   Atlassian CEO Mike Cannon-Brookes told The Verge that the so-called SaaSpocalypse — AI destroying enterprise software businesses — has not played out, even as AI reshapes how companies work. If AI had destroyed enterprise software, the whole category of businesses selling such tools would be at risk — but it has not, according to the head of a company that makes those tools.

   It is not yet clear whether AI tools reading companies' existing systems will make migrating to Atlassian's platform less enticing, or whether AI might build those tools instead.

6. 6

   The AI Conference at Pier 48 draws 120+ speakers

   The AI Conference announced its full agenda, with 120+ speakers and an anticipated 5,000 attendees at Pier 48 in San Francisco from September 29 to October 1. The agenda shifts from hypothetical AI capabilities to real-world deployment, with sessions on what is working, what is not, and how companies are actually building and investing in AI.

   The conference is a paid press release, so attendee turnout and how the applied-AI conversations translate into business decisions are the tests. Watch whether the anticipated 5,000 attendees materialize.