Stocks are tanking for traditional software companies, and the market is calling it a “SaaS Apocalypse.” It seems the era of AI agents eating software interfaces has finally arrived. I just watched this incredible episode from the hosts of Forward Future, who sat down with three industry titans to unpack a wild week in 2026.
🤖 The Battle of the Giants
The hosts broke down two massive, simultaneous releases. First, Anthropic launched Opus 4.6, featuring a one-million token context window and the ability to catch its own mistakes. Minutes later, OpenAI responded with GPT-5.3 Codex. The experts noted a fascinating shift: while Opus still lets you see the code like an IDE, the new Codex acts as a “black box” that just fixes files in the background. Most shockingly, the original creators revealed that GPT-5.3 was instrumental in writing its own code, a huge step toward self-improvement.
⚙️ Breaking the Hardware Lock
The co-founder of Modular argued that the future of AI infrastructure must be open. He explained that developers shouldn’t be locked into one type of chip (like Nvidia’s). Instead, he is building a “hypervisor for compute” that allows models to run seamlessly across any hardware. He also shared a personal workflow where he uses local embedding models to let different AI agents (Claude and Gemini) review each other’s code before merging, drastically cutting token costs!
📉 “Software is Dead”
The investor from Emergence Capital didn’t mince words. He stated that the traditional SaaS business model is over because you can’t just hire sales armies to sell basic software anymore. His firm is now “burning the boats” and pivoting entirely to “physical intelligence,” investing in companies like Bedrock Robotics that build autonomous construction equipment.
🧮 The AlphaGo Moment for Math
Finally, the founder of Axiom shared a stunning achievement: her four-month-old startup built an AI that scored a perfect result on the Putnam exam, the world’s hardest math test. She explained that “math is code,” and by mastering formal verification, we can finally build AI systems that are safe for mission-critical tasks where failure isn’t an option.
🧠 Strategy of the Day: The “Compound Loop”
The expert from Modular shared his personal coding workflow that you can replicate manually:
- Phase 1: Ask Model A (e.g., Claude) and Model B (e.g., Codex) to create a plan.
- Phase 2: Have them review each other’s plans and merge the best parts.
- Phase 3: Generate the code, then have a third model (e.g., Gemini) critique the implementation.
This future is moving fast. Check out the full video to see the new ads Anthropic is running against OpenAI.