🍎 Apple kept the AI agent and shelved the layoffs

Welcome to the Future//Proof 🚀 

👋 Hello , the AI Enthusiast.

In this week’s edition, we brought AI updates backed by high-quality research and data to give you deeper insights. You'll find the Top AI Breakthrough of the Week, a featured AI tool with a mini-tutorial, learning resources to help you master these tools, the top 3 AI news stories, and more.

Our goal is to help you improve your knowledge and stay ahead in the rapidly evolving AI landscape. You can submit your questions, queries, thoughts, opinions or anything regarding AI as a reply to this email and we'll feature and address them in our next newsletter.

🚀 Now Let’s dive in and explore the new AI Insights together!

⌛ Read Time: 5m:35s

An in-depth look at a major AI development, its industry impact, how it could affect your career, and a bold future prediction.

Potential Impact

This is the third week running that a large employer has shown us the gap between AI announcements and AI headcount. AT&T said it expects fewer staff but refused to name a number. Surveys last month found most executives still cannot see productivity gains in the data. Now the company with arguably the best customer support reputation in consumer technology has built the agent, deployed it to live callers and still decided it could not remove the people behind it yet. The pattern is consistent. A bot that contains 70% of calls sounds like 70% fewer agents, until you find that a chunk of those "contained" customers call back angrier, reach a human anyway and take longer to calm down. Containment counts conversations that ended. Resolution counts problems that ended. The distance between those two numbers is where every support automation budget is currently being won or lost, and most dashboards only show the first one.

Implications for People/Careers

If you work in customer support, operations or any queue-based role, Apple has just shown you where your value sits: in the calls the agent cannot close. Get closer to the hard queue, the escalations, the multi-issue cases, the angry repeat callers, because that is the work that survived at Apple. If you manage a support team or are being sold an AI agent, change the question you ask vendors. Stop asking what percentage of contacts the bot handles. Ask what percentage of those customers did not come back within seven days about the same issue, broken down by intent. If the vendor cannot produce that number from your own traffic, you do not have a business case yet. You have a demo.

Our Future//Take

Within a year, "resolution rate" replaces "containment rate" as the headline number in every serious support automation contract, and the layoff announcements that do happen will come after a full quarter of production data, not before. Apple's pause is not a retreat from AI. It is a company refusing to cut staff on a metric it does not trust, which is exactly what yours should do. Your move this week: pull last month's repeat-contact rate for whatever your team automates, even roughly. If nobody can produce it, that is your first project. Here’s your ₹25,000 AI Gift for FREE 🎁 

Quick summaries of this week's top AI news, their relevance to your career, and our expert opinions.

OpenAI's DevDay turned agents into always-on co-workers

At DevDay on September 29, OpenAI shipped a long list, and three pieces matter for working professionals. Dots are always-on agents that keep working in the background on tasks you assign, available on Pro and Business Premium in eligible markets, with Enterprise, Edu and Healthcare in beta. @ChatGPT now works inside Slack and Microsoft Teams for Business and Enterprise customers, alongside new Team Tasks for shared work. And GPT-6.1 Sol arrived, which OpenAI describes as near-Astra intelligence at a fifth of Astra's token prices, strong at coding and computer use. OpenAI also launched Private Intelligence with zero data retention, a Marketplace, and a new Pro 500 tier for heavy users. One day earlier, on September 28, Anthropic released Claude Sonnet 5.5 at the same $2 in and $10 out per million tokens, over 30% faster than Sonnet 5 and scoring within two points of Opus 5.5 on the GDPval-AA work benchmark.

Why It Matters to You

Two things changed. Agents moved from something you open to something that keeps running, and they moved into the chat tools your team already lives in. Meanwhile the two leading mid-tier models now cost exactly the same, which means price is no longer the deciding factor between them. How well each one handles your workflow is.

Our Take

An agent that works while you are offline needs a job description before it needs a login. Pick one recurring task, such as a Monday pipeline summary or a weekly competitor scan, give it to a Dot or equivalent, and review its output every day for two weeks before extending its reach. On the models, run your three most common prompts through GPT-6.1 Sol and Sonnet 5.5 side by side. At identical prices, a one-hour test is the cheapest decision you will make this quarter.

An AI-driven attack wiped cloud resources in seven minutes, using a key left in a GitHub edit history

Microsoft's threat intelligence team published details on September 25 of an attacker it tracks as Storm-3168, which used AI agents to attack Azure environments. The entry point was mundane: a service principal secret pasted in plain text into a public GitHub issue. Someone edited it out, but it remained visible in the edit history. The attackers spent about 15.5 hours quietly reading the environment with over 300 read operations, then ran the destructive phase in about seven minutes, with more than 150 destructive and credential operations inside 35 minutes: over 100 attempts to delete storage accounts, plus Key Vaults, Function Apps and backups. Microsoft observed no ransom note and no confirmed data theft. What survived is the lesson: resources protected by Azure resource locks were not deleted.

Why It Matters to You

Seven minutes is faster than any human on-call rotation can respond. The old assumption that you would notice an attacker and cut them off no longer holds when the attacker is an agent. And note the pattern from this newsletter two weeks ago: a Docker image layer then, a GitHub edit history now. Deleting a secret does not delete it.

Our Take

This is not a task for your security team alone. Ask three questions on Monday. Has every credential ever pasted into a ticket, issue or chat been rotated, not just deleted? Do automated accounts have only the permissions they need? And are your backups and critical resources locked against deletion, even by an admin account? The third question is the one that saved this victim, and it usually takes an afternoon to put in place.

The first US appeals court ruling on AI training went against the AI company

On September 30, the Third Circuit Court of Appeals upheld a lower court ruling for Thomson Reuters against Ross Intelligence, which had used Westlaw's copyrighted legal summaries, known as headnotes, to train an AI legal search tool. Ross argued fair use. The court disagreed, making this the first time a US appeals court has ruled on copyright and AI training. The lower court had found that Ross took the headnotes to build a competing legal research product, so the use was not transformative. The appeals court's reasoning is currently sealed. Ross shut down in 2021, citing litigation costs.

Why It Matters to You

The important limit: Ross's tool found existing documents rather than generating new text, so this does not settle the big generative AI cases. What it does establish is a principle any business can understand. Using someone's content to build a product that competes with theirs is where fair use is weakest.

Our Take

If your company is training or fine-tuning a model, or building a retrieval tool, on data you did not create, list the sources this week and mark any that come from a company whose product yours could replace. That column is where your legal risk now concentrates. For everyone else, expect more AI vendors to advertise licensed training data as a selling point, and ask your vendors where their data came from before your legal team does. AI Mastery for FREE (Sign up Now).

Discover a comprehensive guide to an AI tool, exploring its features, practical use cases, and learning resources to help you master it.

🔗 ElevenLabs Eleven v4

ElevenLabs released Eleven v4 and Eleven v4 Turbo on September 28, and it is the first voice model where you can direct the performance the way you would brief a voice artist. It speaks 90+ languages, and a voice built in one language keeps its identity when it speaks another. You steer delivery in plain English or with inline tags like [laughs] or [whispers], and in multi-speaker scripts each voice responds to the line before it. v4 Turbo is built for live voice agents, with a median of about 150 milliseconds to first speech. For Indian businesses selling across states or abroad, this is the week one recorded voice became a multilingual one.

⭐ Top Features

• One voice, many languages. Record your founder, trainer or brand voice once and have it deliver in Hindi, Tamil, English or Spanish while still sounding like the same person.

• Direction in plain English. Tell it "calm and reassuring, slight smile" instead of fighting with settings.

• Inline performance tags. Drop [pause], [laughs] or [whispers] into the script exactly where you want them.

• Natural multi-speaker dialogue. Voices react to each other, which makes training role-plays and podcast-style explainers sound far less robotic.

• v4 Turbo for live agents. Fast enough for phone and support agents, available in ElevenAgents and the API.

Resources for Learning

• Official Launch Post: elevenlabs.io/blog/eleven-v4 for features, tags and availability.

• Independent Coverage: TechCrunch for a plain summary of what changed.

A curated list of noteworthy AI tools and their key details to help you stay ahead in your field.

Manus 2.0

Launched September 28, this is Manus rebuilt as a full work platform. A new engine called Cascade uses about 23% fewer tokens, 28% less time and 32% less cost per task. A Cloud Computer keeps running when your laptop is closed, Automations start work from an email, calendar event or Slack message, and Manus Studio produces documents, spreadsheets, websites and videos in one place. The differentiator is event-triggered agents: work begins when something happens, not when you remember to ask. Caveat: pricing for the new features was not published at launch.

Fo by Wajo

Publicly launched September 28 by former Google DeepMind engineering lead Shivani Poddar. Fo is a personal agent with its own phone number, email address and payment card, so it can call a clinic, book the appointment and pay the deposit. When a task gets hard, it hands off to a real human assistant who sees only that task. In Wajo's own benchmark it finished 71% of real errands without help. The differentiator is that it acts on the phone, which is still how most of the world's small businesses take bookings. Caveat: US adults only for now, and the benchmark is the company's own.

Base Code by Base44

Launched September 28 by Wix-owned Base44. Base Code connects your existing GitHub codebase to a shared cloud workspace where product managers, designers, QA and marketing can request changes through AI chat, see a live preview, and submit the result as a pull request for engineers to approve. The differentiator is that it works on the software you already have, not a new app built from scratch, while keeping developers in control of what ships. Caveat: no pricing was announced, and letting more people touch production code only works if your review process is actually enforced.

Tapkit

Introduced September 28, Tapkit is a Mac app that lets AI agents, including Claude, see, tap and type on a real iPhone, using Apple's accessibility features. That means an agent can work inside apps that have no API, such as a supplier's ordering app or a regional marketplace. It costs $49 a month per phone and you bring your own device. The differentiator is reach: much of India's commerce runs on mobile-only apps that desktop agents cannot touch. Caveat: use a dedicated phone, never your personal one, because the agent sees everything on it.

A quick poll to help you recollect and engage with key points from the newsletter.

Share your feedback on today's edition to help us improve and better meet your needs.

Share our Newsletter ⏩

Enjoying our insights on the latest AI breakthroughs? Don’t keep it to yourself! Share this newsletter with friends and colleagues who are passionate about technology and AI innovation.

If you haven’t subscribed yet, make sure to subscribe here to stay updated with cutting-edge AI news, tools, and tutorials delivered straight to your inbox!

Ask Us Anything AI ❓

Got questions? We've got answers!

Submit your questions, queries, thoughts, opinions or anything regarding AI and we'll feature and address them in our next newsletter. Your curiosity drives our content!

👇 Reply to this email with your questions, and we'll answer them in our next edition!👇