Better AI Starts With Better Prompts
Most people type short prompts. The smartest AI users speak everything they mean.
Flow captures your full thoughts, removes filler words, and creates polished prompts that produce richer, more accurate AI outputs - all for free.
Today In Ai
1 Apple raised MacBook and iPad prices by up to $400, and AI is directly to blame.
AI data centers are consuming memory chips at a pace that has left almost nothing for consumer devices, and Apple is now passing that cost directly to buyers. Price increases hit Macs, iPads, and HomePods on Thursday, with the 13-inch MacBook Air jumping from $1,099 to $1,299 and the base iPad climbing from $349 to $449. Apple said it has never seen a component price increase this large, this fast.
Memory prices have more than doubled since October 2025. Analysts project another 30-40% climb this year, and Micron expects the shortage to last through 2027 at minimum. Apple shares fell 6% Thursday. Tim Cook steps down September 1, handing this mess to incoming CEO John Ternus on day one. Analysts expect iPhone Pro prices to follow later this year, potentially rising $200.
2 OpenAI leans toward delaying its IPO to 2027 after SpaceX's rocky debut spooked its advisers.
Sam Altman wants a $1 trillion valuation. His CFO reportedly wants to wait until that number is actually achievable. The pressure to hold off grew after SpaceX went public two weeks ago, briefly touched $2 trillion, then fell more than 25% from its peak. OpenAI is now weighing a 2027 target, with the company burning $3.7 billion per quarter and the market mood shifting fast.
3 Google bakes computer use directly into Gemini 3.5 Flash.
Developers can now build agents that see, click, and control browsers, mobile apps, and desktop software without routing through a separate model. Computer use is now a standard capability inside one of the fastest and cheapest frontier models available, which dramatically lowers the cost of building agents that interact with real software in the real world.
Bonus reads: RAISE US launched with $500M from OpenAI, Anthropic, Amazon, Microsoft, and Bank of America to retrain workers displaced by AI, starting in four states. Adobe acquired Topaz Labs, bringing its AI sharpening and upscaling models into Photoshop, Lightroom, and Premiere.
From The Frontier
How we got here. Every AI data center being built right now runs on the same memory chips inside your MacBook, iPad, and iPhone. Samsung, SK Hynix, and Micron, the three companies that control global memory production, can only make so many. When hyperscalers started ordering at scale, the price of DRAM more than doubled in eight months. Consumer electronics manufacturers, Apple included, had no leverage to compete on price.
Why Apple blinked first. Most hardware makers absorbed the cost quietly for as long as they could. Apple moved because the gap between component cost and shelf price became impossible to close at volume. The company's own statement, that it has never seen a component price increase this large, this fast, is not marketing language. It is the kind of sentence that gets written when the finance team runs out of other options.
Who wins from this. The three memory producers are having their best period in years. OpenAI's Codex user base grew more than 5x in the first half of 2026, with non-developer adoption rising 137x for individual users, which means AI demand for compute and memory is accelerating, not leveling off. The shortage is not a blip. It is a structural consequence of building an entirely new layer of global infrastructure in a very short window of time.
The forward view. New chip factories take three to five years to come online. The analysts projecting another 30-40% memory price increase this year are not being alarmist. They are reading the construction permits. If you have been waiting to upgrade your laptop or phone, the window that existed six months ago has closed. The next one opens sometime around 2028, when new supply finally catches up to what AI built.
What people are actually watching and sharing
Codex goes mainstream. The detail buried in OpenAI's Codex growth report is the one that matters most: legal, finance, and recruiting teams at OpenAI itself now use Codex as their primary work tool, not ChatGPT. The full breakdown of how agents are transforming work internally is worth reading as a preview of what enterprise adoption looks like at the frontier.
GPT-5.6 gets a slower rollout. The Trump administration asked OpenAI to stagger access to its upcoming GPT-5.6 model while the government approves customers one by one over security concerns. OpenAI agreed, meaning the model will launch in a limited preview to select partners before any broader release. The era of same-day public access to every new frontier model may be ending.
AI coding talk worth 40 minutes. Developer Aleksander Stensby's NDC Copenhagen session on getting better results from AI coding tools has been circulating among engineering teams this week. The three core techniques he covers apply whether or not you write code: give the AI full context, define what done looks like, and iterate rather than restart.
Tim Cook's exit timeline. Cook steps down as Apple CEO on September 1, handing the role to John Ternus, who will inherit a memory shortage, a stalled Vision Pro category, and an AI hardware story that still needs to be written. The timing is not ideal for a new CEO trying to establish credibility with investors.
Prompt Station
Fix broken code without starting over from scratch
Most people respond to a failed code output by deleting everything and trying a completely different prompt. Developer Aleksander Stensby's NDC Copenhagen talk makes the case for a better approach: describe exactly what is wrong, specify exactly what to fix, and tell the AI what not to touch. This template, adapted from his session, works in Claude Code, ChatGPT, and any AI coding tool.
COPY AND PASTE THIS PROMPT
I'm working on [WHAT YOU'RE BUILDING]. Here's my current code:
[PASTE YOUR CODE HERE]
The issue is: [SPECIFIC PROBLEM OR EXACT ERROR MESSAGE]
Please fix [SPECIFIC THING TO FIX], making sure it [REQUIREMENT 1], [REQUIREMENT 2], and [REQUIREMENT 3]. Don't change anything else.Fill in each bracket with specifics. For [WHAT YOU'RE BUILDING], try: "a user login form in React", "a Python script that parses CSV files", or "a Shopify checkout flow". For [REQUIREMENT 1-3], think about what good output looks like: handles empty inputs, includes error messages the user can read, and doesn't break the existing styles. The more precisely you describe the end state, the less the AI has to guess. Source: Aleksander Stensby, NDC Copenhagen.
Your best prompts are the ones you'd never bother typing.
The detailed ones. The ones with examples and edge cases. Wispr Flow lets you speak them instead — clean, structured, ready to paste into any AI tool. Free on Mac, Windows, and iPhone.