August 5, 2026
|
|
Greetings, this week: OpenAI cuts the price of its cheapest model by 80% just three weeks after launching it, Microsoft folds every Copilot into one app and quietly reveals 30 million paid seats, AI-generated noise buries a real $200,000 security fix at Apple, a five-step way to build the internal tool your IT queue keeps postponing, and Cohort 2 of our AI Accelerator for Individuals opens ahead of an August 15 start.
|
|
In Today's Signal
|
| Signal |
OpenAI cuts its cheapest model's price by 80%, three weeks after launch |
|
| AI News |
Microsoft merges every Copilot into one app, and AI slop drowns Apple's bug bounty |
|
| Steal This |
Build a working internal tool without writing one giant technical prompt |
|
| Futurelab |
AI Accelerator for Individuals: Cohort 2 starts August 15 |
|
|
|
| Signal of the Week |
 |
OpenAI cut its cheapest model's price by 80%, three weeks after launching it
| On July 30, GPT-5.6 Luna dropped from $1 to $0.20 per million input tokens, and $6 to $1.20 on output. Mid-tier Terra fell 20% to $2 and $12. The flagship Sol was left alone at $5 and $30. |
| The repricing came just three weeks after the GPT-5.6 family went generally available on July 9, which is fast enough to be a reaction rather than a plan. |
| OpenAI says its own Codex tool rewrote the production GPU code that runs these models, cutting serving costs by roughly 20% and funding part of the discount. |
| CNBC read the cut as a direct response to companies turning cost-sensitive about AI, after a year of buying on capability and asking about the bill later. |
Futurelab Take
Nobody cuts 80% off a three-week-old product from a position of strength. The volume tier is where enterprise money actually sits, and cheap open models were taking it. If you buy AI at scale, your negotiating position just improved without you doing anything.
Source: "OpenAI cuts prices for two of its GPT-5.6 AI models as companies grow sensitive to costs, Image Source: OpenAI"
|
|
| AI News That Matters |
| Enterprise Adoption |
 |
Microsoft is collapsing every Copilot into one app, and revealed 30 million paid seats
| On the July 30 earnings call, Satya Nadella confirmed a single app merging Copilot chat, GitHub Copilot, Copilot Cowork and the Autopilot agent system, for consumers and businesses alike. |
| He described the arc as chat, to Cowork, to Autopilots, with the pieces including code coming together this quarter. No product name and no launch date were given. |
| Buried in the same call: Microsoft 365 Copilot has now passed 30 million paid seats, which makes it one of the fastest-scaling paid software products the company has ever shipped. |
| The same week, Azure crossed $100 billion in annual revenue and Microsoft added roughly $450 billion in market value in a single day. |
Futurelab Take
Thirty million paid seats is the real headline. Assistants have moved from pilot budgets to standing licence lines. And the merger tells you what Microsoft learned selling them: too many overlapping AI tools confuses buyers. That is the single most common reason internal rollouts quietly stall.
Source: "Microsoft to Merge Copilot, GitHub Copilot, and AI Agents Into One App"
|
| AI Slop and Broken Systems |
 |
AI-generated noise flooded Apple's bug bounty, and a $200,000 fix never got through
| Reported August 3: Apple's bug bounty programme, which pays researchers to report security flaws, has been swamped by AI-generated vulnerability submissions. |
| The backlog hit capacity, and a genuine macOS vulnerability that Apple was willing to pay $200,000 for never got reported. The noise did not just annoy anyone, it cost a real fix. |
| The programme was designed around a quiet assumption: that writing a convincing security report takes real human effort, so volume would stay within human reach. That assumption no longer holds. |
| The same design flaw sits inside job applications, RFP responses, peer review, grant panels and support queues, every intake process built for human-scale input. |
Futurelab Take
This is the clearest case study yet of AI slop doing measurable damage rather than causing mild irritation. Every intake process you run was sized for a world where producing input cost something. That world is gone. Redesign the filter before it drowns, not after.
Source: "Apple Bug Bounty Overwhelmed by AI-Generated Submissions"
|
|
| Practical Tool - Steal This |
 |
| Internal Tools - ChatGPT plus Lovable |
Build a working internal tool without writing one giant technical prompt
The pain: the tool you need does not exist, and the IT queue for it is six weeks long.
| 1. | Open ChatGPT, attach a screenshot or sample of the thing you want, and describe the tool in plain English. Let it ask you a few questions. |
| 2. | Keep version one to one journey only, for example: paste in data, convert it, review it, save it. |
| 3. | Ask ChatGPT to convert the conversation into concise feature requirements plus a brand brief for Lovable, covering colours, feel and typography. |
| 4. | Delete every extra feature from that handoff before you copy it. This is the step most people skip. |
| 5. | Create a Lovable project, attach the same screenshot, paste the handoff, then enable Lovable's built-in AI under Connectors and test it. |
Copy-paste prompt
"Turn our conversation into concise feature requirements and a brand brief for Lovable. Include colours, feel and typography. Cover one user journey only. No extra features, no nice-to-haves."
Why it works: most no-code builds fail because the first prompt tries to describe the whole product. Getting a clean, narrow spec written first is what makes the build work.
Tools: ChatGPT plus Lovable. The free tier is enough to test the idea.
|
|
| From Futurelab |
 |
|
Cohort
AI Accelerator for Individuals: Cohort 2 starts August 15
Four weeks, live, designed to move you from casual prompting to workflows you actually keep using. You will work hands-on with Claude Cowork, Codex, ChatGPT, Perplexity Computer and NotebookLM, and learn where each one fits and where it fails. Every exercise runs on your own real work, not sample tasks, so you leave with three to five reusable workflows plus one built around a problem you are dealing with right now. No coding needed. Cohort 2 starts August 15, and seats are limited.
|
|
| Join the Conversation |
5,000+ professionals
Founders, business leaders, and practitioners navigating the AI transition together. No noise. No self-promotion. Just people serious about getting it right.
Join the Community
|
|
© 2026 Futurelab Studios. All rights reserved.
To unsubscribe please click here.
|
|