October 14

Your Daily Note,

AI is moving out of the chat window and into factories, phones and the systems that keep businesses running. Robots are learning to switch tasks without weeks of reprogramming. Google is making it easier to build personal apps with AI. And security companies are trying to keep up with software that employees can create faster than traditional security teams can review it.

In today’s Noted:

Factory robots could switch tasks in minutes, not weeks

We noted:

Industrial robots are good at repeating the same task thousands of times. The problem is that traditional automation can be difficult to adapt when a factory changes its product, production line or assembly process. Reprogramming and physically retooling a line can take weeks or months.

Industrial Next wants to make factory robots more flexible with AI models that connect visual perception, an understanding of the task and physical movement. Instead of programming every movement by hand, manufacturers could teach a robot new tasks through simulation and demonstrations.

The company says one of its systems can remove protective film from phone camera modules and buttons at human-level or faster speeds, with 100% accuracy in its described tests, including after moving to an unfamiliar phone model and production line. That is a company-reported result, not a guarantee across all factory tasks.

Why it matters

If the approach works across more production environments, manufacturers could adapt production lines more quickly when a product changes. Smaller production runs and more varied products may also become easier to automate.

When evaluating industrial AI, look for evidence that a robot can handle a new task or unfamiliar part, not just repeat a carefully prepared demonstration.

Google is making it possible to build an app just for yourself

We noted:

AI coding tools have made it easier for people without traditional programming experience to create small applications. But turning a prototype into something that runs natively on a phone has often required extra setup or publishing it through a public app store.

Google is changing that with support for private, AI-built Android apps through Google Play. Users can create a personal app with Google AI Studio and install it on their own Android device without making it publicly searchable.

The initial rollout is limited to users aged 18 and older in the United States. It is not yet a general release for every Android user worldwide.

Why it matters

Imagine creating a simple household planner, a personal reading tracker or a small tool that organizes information in exactly the way you prefer. Instead of searching for an app that almost fits, you could describe what you need and build a custom version.

For small businesses, the same approach could help create internal tools for specific tasks without commissioning a full software product.

Start with a small app that organizes information or automates a low-risk task. Before connecting it to sensitive accounts or customer data, check what information it stores and what permissions it requires.

AI is creating security problems faster than companies can review them

We noted:

Employees can now build small apps, install AI extensions and connect agents to company systems without waiting for a traditional software development cycle. That speeds up experimentation, but it also creates a new problem: businesses may not know what their employees have built, what those tools can access or whether their components are safe.

Pluto Security has emerged from stealth with $42 million in funding to help companies monitor and control their use of AI tools and agents.

The company says roughly 60% of the risks it detects across its customer base involve tools built by non-developers using AI. It also reports more than 30 enterprise customers and over $2 million in annual recurring revenue. These are figures provided by the company and reported by TechCrunch.

Why it matters

Businesses adopting AI may increasingly need tools that track not only what employees do, but also what their AI agents and custom-built apps are doing on their behalf. The aim is to allow employees to experiment with AI while giving security teams visibility into the systems being created and the data those systems can access.

Before rolling out AI across a team, establish which tools are approved, what data they can access and who is responsible for reviewing custom-built apps.

AI video generation is moving beyond one-minute clips

We noted:

Higgsfield AI is using OpenAI models to push AI-generated video into longer formats. The company says that before its latest workflow, its generated videos typically topped out at around one minute. With GPT-6 Astra, it says creators can now produce videos longer than 10 minutes.

The goal is not just to generate a longer clip. It is to maintain visual consistency across a sequence of scenes while reducing the amount of manual work needed to produce a finished video.

The longer-video capability and production improvements are described in OpenAI’s customer case study. Results will vary by workflow, and generating a longer video does not automatically make every scene consistent or ready for publication.

Why it matters

Small creative teams could use these tools to produce longer explainers, tutorials, product demonstrations and marketing videos without coordinating every production step manually. For businesses, the benefit may be the ability to test more creative ideas at a lower production cost. Human review will still be important for factual accuracy, branding, continuity and rights.

When testing an AI video tool, judge the finished sequence rather than a single impressive frame. Look for consistency across scenes, accurate text, coherent pacing and how much editing remains.

AI is being used to stress-test robots before they enter the real world

Image source: Laplace

We noted:

When an AI system controls a physical machine, a mistake can have consequences beyond a bad answer or a broken workflow. A robot may encounter an unusual object, a self-driving vehicle may face an unexpected road situation, or a system may behave differently outside its training environment.

Laplace Intelligence is building automated testing for physical AI. It uses simulated environments and AI agents to search for situations in which robots and autonomous vehicles might fail.

Laplace is building on research from the University of Michigan. The company’s approach is to use AI to challenge AI, exposing weaknesses before systems are deployed.

Why it matters

Developers of autonomous vehicles, warehouse robots and other physical AI systems could use automated testing to discover more failure cases before real-world deployment. The value will depend on how accurately the simulation represents reality and whether the scenarios it generates correspond to failures that can occur outside the simulator.

For physical AI, ask how a system was tested, which edge cases it was evaluated against and what real-world evidence supports its safety claims.

Products worth trying

 

Everything else in AI

Facebook lets users tell its algorithm what they want to see

Meta has launched Your Algorithm on Facebook, giving users a way to describe in their own words what they want to see more or less of in Feed and Reels.

Rather than relying entirely on likes, watch time and other behavioural signals, users can directly express preferences. Meta says recommendations can update in real time, and the three-dot menu on posts makes it easier to indicate interest or disinterest. The feature builds on similar personalization controls on Instagram and Threads.

Why it matters: Recommendation algorithms have traditionally inferred what people want from their behaviour. Natural-language controls give users another way to explain their preferences directly. The real test will be whether those controls consistently change the feed in ways people find useful.

 

Flox Intelligence uses AI to keep wildlife away from transport routes

Wildlife can cause costly disruption to roads, railways and airports, while collisions can also harm animals. Flox Intelligence is building an alternative to relying only on fences, sirens and other fixed deterrents.

Its devices use motion sensors and cameras to identify nearby animals, then emit sounds tailored to the species and situation. The company says the system varies its responses so animals do not simply become accustomed to the same signal. Flox reports around 150 devices deployed across North America, Europe and Australia. It has partnered with train manufacturer Alstom and is opening a $10 million Series A round to expand deployments.

Why it matters: This is an example of AI perception being applied outside familiar digital environments. The system must interpret a changing natural world and respond in a way that protects both infrastructure and wildlife.

 

Liquid AI releases a 24-billion-parameter model designed for on-device agents

Liquid AI has released LFM2.5-24B-A2B, an open-weight reasoning model with 24 billion total parameters but only around 2 billion active for each token.

The company says this design helps the model run faster than comparable mixture-of-experts models on local hardware, making it more suitable for private, on-device agents. Liquid AI reports speeds of 184 tokens per second on an Apple M5 Max and says the model performs competitively on several agent and tool-use benchmarks. The model is available through Hugging Face and Liquid AI’s Playground, along with a base version intended for further development.

Why it matters: Bigger models are not always the best fit for everyday applications. Models designed to use less computing at inference time could make local AI agents more responsive and reduce dependence on cloud services.

 

Cerebras CEO pushes back against AI doomsday warnings

Cerebras CEO Andrew Feldman argues that public discussion of AI risks is sometimes overshadowing the technology’s potential benefits, including advances in healthcare and scientific research.

He does not argue that powerful AI is risk-free. Instead, he says the industry needs to acknowledge both the potential benefits and the dangers, regulate the technology reasonably and improve its relationship with communities affected by data-centre expansion. Feldman has a commercial stake in AI adoption: Cerebras sells computing hardware to AI developers. His comments are therefore one perspective in a wider debate that includes researchers and security experts who warn about increasingly capable systems.

Why it matters: AI risk debates often become framed as a choice between enthusiasm and alarm. The harder question is how to capture useful benefits while developing evidence-based safeguards for genuine risks.

 

A judge orders OpenAI to hand over hiring records in Apple’s trade-secrets case

A US judge has ordered OpenAI to provide Apple with certain hiring records and interview materials connected to former Apple employees who joined OpenAI’s consumer hardware and robotics teams.

The documents could include interview guides, technical presentations and records of discussions about Apple technology. The order also allows certain depositions and access to private messages involving people named in the case. The judge did not grant every request Apple made. In particular, the court declined to expedite access to some broader hardware research and development materials, while leaving room for narrower requests later.

Why it matters: The dispute highlights the legal and competitive pressures surrounding AI hardware. As companies compete to build new devices, employee mobility, confidential information and product development are becoming part of the same battle.

Worth your weekend

See you tomorrow
same time, different note.

~ Stackari

 

P.S. Before you connect any AI tool to company systems, check what it can access.