EPISODE · WEDNESDAY, MARCH 25, 2026

Siri Becomes an Agent, OpenAI Kills Sora, and Why 97% of AI Content Fails After 3 Months

Apple is testing a standalone Siri app for iOS 27 and macOS 27 that transforms Siri from a voice assistant into a system-wide AI agent — unveiling expected at WWDC June 8. OpenAI has shut down Sora just months after launch; the real reason is compute economics: $13B revenue vs $100B projected spend. A 16-month experiment shows 97% of AI-generated pages fell out of the top 100 after 3 months without authority signals. Plus: Claude Code auto mode, MiniMax M2.7 engineering agent, the SaaS revenue durability crisis, and the end of free trials.

Ready to Play

0:00 / 9:10

In This Episode

  • • Apple testing standalone Siri app for iOS 27 / macOS 27 — system-wide AI agent with deep platform integration
  • • WWDC June 8 unveiling — 10 weeks to close the gap on Google, Anthropic, and OpenAI agents
  • • OpenAI shuts down Sora — compute economics: $13B revenue vs $100B projected 4-year spend
  • • Disney already shopping for alternative AI video partners post-Sora shutdown
  • • OpenAI consolidating into a superapp — Sora shutdown is first visible product cut
  • • Claude Code auto mode (research preview) — safe actions proceed automatically, risky ones blocked
  • • Anthropic direction: ask less, execute more autonomously
  • • 16-month AI content experiment: 71% indexed in 36 days, 526K impressions by month 3
  • • 97% of AI pages fell out of top 100 after 3 months — authority signals are the differentiator
  • • 70% of AI content traffic came in first 2.5 months, then collapsed
  • • MiniMax M2.7: 9 gold medals on MLE-bench, 56.22% SWE-Bench Pro — small teams run complex engineering
  • • SaaS revenue durability crisis: Bill.com 90%→12%, Snowflake 73%→26% growth
  • • Free trials declining — paid trials and qualifier events becoming the new standard

Show Notes

  • Introduction — Siri agent, Sora shutdown, 97% AI content failure
  • Apple Rebuilds Siri — standalone app, iOS 27, WWDC June 8, system-wide agent
  • OpenAI Kills Sora — compute economics, $13B vs $100B, Disney shopping alternatives
  • Claude Code Auto Mode — safe/risky classification, autonomous execution, research preview
  • The 97% Problem — 16-month AI content experiment, authority signals, cluster content
  • Quick Hits — MiniMax M2.7, SaaS revenue durability crisis, end of free trials
  • Actionable Items — audit AI content, title tag clarity, agent interview, trial structure
  • Outro

Transcript

[0:00] Introduction

Welcome to the AI Daily Digest for Wednesday, March 25th, 2026. I'm your host, and today's episode covers three stories that each challenge something the industry thought it understood.

Apple is quietly rebuilding Siri from the ground up — not as a voice assistant, but as a system-wide AI agent. OpenAI has shut down Sora, its video generation platform, just months after launch, and the reason tells you a lot about where frontier AI economics are heading. And a 16-month content experiment has produced a number that should concern anyone relying on AI-generated content for organic search: 97% of AI pages fell out of the top 100 after three months.

We also have Claude Code's new auto mode, a new engineering agent model that scored nine gold medals on MLE-bench, and the quiet death of the free trial in SaaS. Let's get into it.

[0:45] Apple Rebuilds Siri as a System-Wide AI Agent

Apple is testing a standalone Siri app for iOS 27 and macOS 27, with an 'Ask Siri' feature that transforms it from a voice assistant into a system-wide AI agent with deep platform integration. The unveiling is expected at WWDC on June 8th.

This is a significant strategic shift. The current Siri is a voice interface bolted onto Apple's operating system. The new Siri is being designed to operate across the entire platform — reading context from apps, taking multi-step actions, and acting autonomously on behalf of the user.

The competitive pressure here is obvious. Google's Personal Intelligence is now free for all US users. Anthropic's Claude Computer Use is on macOS. OpenAI's Operator is in market. Apple has been visibly behind on agentic AI, and a standalone Siri app with deep OS integration would be the most credible response it could offer.

If the standalone app ships with iOS 27 in September, Apple will have closed most of the gap with Google and Anthropic in the consumer agent space. If it slips, the competitive window narrows further.

[1:35] OpenAI Kills Sora: What the Compute Bill Actually Means

OpenAI has shut down Sora, its video generation platform, just months after launch. The official reason is that compute resources are being redirected toward productivity tools ahead of a potential IPO. The real reason is the economics.

OpenAI made approximately 13 billion dollars last year. It expects to spend approximately 100 billion dollars over the next four years. Video generation consumes dramatically more compute than chat — a single video generation job can cost orders of magnitude more than a text response.

Disney, which had been working with Sora, is already shopping for alternative AI video partners. The broader signal is about product consolidation. OpenAI is reportedly planning to merge its products into a single superapp. Sora's shutdown is the first visible step in that consolidation.

[2:25] Claude Code Auto Mode: Anthropic Pushes Toward Autonomous Execution

Anthropic has launched auto mode for Claude Code, currently in research preview. In auto mode, Claude decides which actions are safe to take autonomously, proceeds with those automatically, and blocks actions it classifies as risky.

This connects to a broader strategic direction: Anthropic wants Claude to ask less and execute more. Auto mode classifies actions into safe and risky categories, executes the safe ones without interruption, and only surfaces the risky ones for human review.

The research preview status means this is not yet production-ready, but the direction is clear: autonomous execution with selective human oversight, rather than human approval at every step.

[3:10] The 97% Problem: Why AI Content Fails Without SEO Fundamentals

A 16-month content experiment produced a finding that deserves serious attention. 71% of AI pages were indexed within 36 days. By month three, those pages had grown to 526,000 impressions. But only 3% remained in the top 100 after three months. 70% of all traffic came in the first two and a half months, and then collapsed.

AI content that lacks authority signals, unique insights, and trust signals gets a short-term indexing boost, but then gets displaced by pages with stronger signals once Google's quality evaluation catches up.

The actionable implication: treat AI as a drafting tool rather than a publishing tool. Every AI-generated page needs a layer of unique insight, original data, or expert perspective. The pages that held their rankings combined AI efficiency with human authority signals.

[3:55] Quick Hits: MiniMax M2.7, Revenue Durability Crisis, End of Free Trials

MiniMax M2.7 is a new cowork agent model built for real engineering work across 50+ skills. It scored nine gold medals across 22 MLE-bench tasks and 56.22% on SWE-Bench Pro. Small teams can now run complex engineering workflows without expanding headcount.

VCs and PE firms are losing confidence in software revenue durability. Bill.com dropped from 90% to 12% growth. Snowflake from 73% to 26%. AI is disrupting the workflows that SaaS tools were built to support.

Free trials are declining. AI-driven costs and shorter payback expectations are pushing companies toward paid trials and trial qualifier events.

[4:35] Actionable Items

Four things worth acting on today.

First, audit your AI-generated content for authority signals. If you have pages that got an initial traffic spike and then dropped, add original data, expert quotes, or unique case studies.

Second, review your title tags for clarity over creativity. Specificity beats cleverness across Google search, AI tools, and social link previews.

Third, try the agent interview pattern: 'Interview me to get all the information you need for this task.' It overcomes blank-page syndrome and captures preferences the model would otherwise miss.

Fourth, if you are a SaaS founder, review your trial structure. Paid trials with a clear value demonstration event are becoming the new standard.

[5:10] Outro

That is the AI Daily Digest for Wednesday, March 25th, 2026. Siri going agentic, Sora shutting down, and the 97% AI content failure rate — three stories that each reveal something important about where the industry is right now.

If you found this useful, subscribe at homepage.marketing and share it with someone who would benefit. We will be back tomorrow with the next digest. See you then.

Related Episodes

More episodes you might enjoy based on similar topics and categories.

AI Models & Research

Tuesday, March 24, 2026

Claude Takes Control of Your Mac, OpenAI's Capital Crunch, and Why Agent Interviews Beat Single Prompts

Anthropic's Claude Computer Use lands on macOS for Pro/Max users — full keyboard, cursor, browser, and file control with permission safeguards. OpenAI is offering private equity 17.5% guaranteed returns, a signal that frontier AI economics are harder than the public narrative suggests. The agent interview technique produces dramatically better AI output than single prompts. Plus: 53% of agency owners see AI as a significant threat, Google Personal Intelligence goes free for all US users, and ChatGPT citation data confirms cluster-based content wins.

AnthropicOpenAIAgentsMarketing

8:13

AI Models & Research

Wednesday, March 18, 2026

GPT-5.4 Mini Costs Less Than a Stamp, Google Now Reads Your Gmail, and Why Faster Code Is Making Teams Slower

OpenAI releases GPT-5.4 mini ($0.75/M tokens) and nano ($0.20/M tokens) — production models designed for high-volume workflows. Google expands Personal Intelligence across Search, Gemini, and Chrome, pulling from Gmail and Photos to personalise results. AI overviews cut click-through rates by 58%, but AI-referred visitors convert at higher rates. Stripe's coding agents push 1,300 PRs/week using blueprint hybrid workflows. And the Atrophy Paradox: heavy AI use reduces cognitive engagement and makes output converge.

OpenAIGoogleSEOMarketingCoding

7:10

AI Models & Research

Friday, March 13, 2026

GPT-5.4 Cites Brand Sites 56% of the Time, Google Maps Gets AI, and the Responses API Changes Everything

GPT-5.4 now cites brand sites 56% of the time — up 7x from GPT-5.3. Pricing pages cited 34x more. Google launches Ask Maps with Gemini. OpenAI Responses API gives agents full computer access. Meta's Avocado model delayed to May. Anthropic study: AI impacts knowledge workers most.

OpenAIGoogleSEOGEOMeta

5:48

AI Models & Research

Thursday, March 5, 2026

Gemini 3.1 Flash Lite Beats GPT-5 Mini, Claude Code Goes Voice, and Bing Makes GEO Official

Google launches Gemini 3.1 Flash Lite — 2.5x faster than previous version, outperforming GPT-5 mini and Claude 4.5 Haiku at $0.25/1M tokens. Claude hits $2.5B run-rate and overtakes ChatGPT in App Store. Apple's MacBook Neo launches at $599 with A18 Pro chip. Bing officially adds GEO to webmaster guidelines with opt-out mechanism. DeepSeek V4 drops with 1 trillion parameters and 1M context window.

GoogleAnthropicOpenAILLMsResearch

8:10