The "agents will kill software companies" thesis took a beating this week. Atlassian just posted a huge quarterly beat, and the argument for why should sound familiar if you've been reading this newsletter: the more agents do, the more governance, workflow, and data management matter. More code being written means more code to track. The platforms don't disappear. They get load-bearing.
01
Atlassian's big quarter quietly buries a bad take about AI
Box CEO Aaron Levie called out what he sees as a misread that's been circulating for months: the idea that AI agents would cannibalize enterprise software categories. Atlassian's results are his exhibit A. When agents are generating 100x more code and making decisions across company systems, the tools that manage that work, track governance, and connect workflows become more critical, not redundant. Levie is essentially arguing that AI creates more surface area for enterprise software, not less. ---
Why it matters: If your company paused renewals on project management or workflow software on the assumption that agents would replace them, this quarter is worth reconsidering. The governance layer isn't going away. It's probably getting a budget increase.
OpenAI just gave GPT-5.6 Luna to everyone, for free, with no cap
ChatGPT free users now get unlimited text conversations powered by GPT-5.6 Luna, the same model developers have been paying to access via API. No subscription required. ---
Why it matters: Yesterday we noted someone was getting rate-limit reset requests every six minutes. OpenAI apparently decided the answer was to remove the limit entirely. At 1 billion users, the only thing left to compete on is whether people actually trust the product with their data and apps.
For coding tasks, DeepSeek costs 4.8x less than GPT-5.6 Luna, and it's not close
Together AI ran 900 coding benchmark rollouts comparing DeepSeek-V4 Flash and GPT-5.6 Luna. Luna wins on raw performance by 14 points on the pass-at-first-try metric. DeepSeek delivers 4.8x more successful solves per dollar spent. ---
Why it matters: If you're running an AI coding pipeline at any volume, this gap means DeepSeek is the default choice unless you absolutely need the accuracy ceiling. The "best model" conversation and the "right model for production" conversation are increasingly different conversations.
Consumer AI is OpenAI and Google's race to lose, and trust is the only real obstacle
Product builder Peter Yang argues that the consumer AI market is already effectively decided: ChatGPT has 1 billion users, OpenAI is ahead on voice and new interaction patterns, and the main job now is getting those users to actually connect their apps and let agents act on their behalf. He pins the two main blockers as distrust of giving AI access to personal data, and most people simply not knowing what ChatGPT can do now. ---
Why it matters: The companies trying to build consumer AI apps on top of someone else's model are competing for the users OpenAI hasn't yet convinced to upgrade, not for some untapped market. That's a much smaller addressable opportunity than it looks.
Madhu Guru shared the workflow he uses with his teams: record yourself explaining an idea out loud as you would to a friend, use AI to do basic cleanup while preserving the original structure, then share that. His observation is that something gets lost between a person's mouth and a polished document: the core idea gets buried under context and polish.
Why it matters: If your team's memos read like they were written to impress rather than explain, the problem usually isn't the writing. It's the sequencing. Voice first, document second.