Cognitive warmup. There is worrying data for artificial intelligence (AI), enterprises and AI in enterprises. Adobe’s latest 2026 State of AI Readiness Report suggests that only 6.7% of Indian enterprises have scaled AI across multiple workflows, while 59% are struggling to move beyond AI pilots to enterprise-wide implementation. This is despite 77% of Indian companies increasing AI investments year-on-year, and the end result is that only 15 out of every 100 AI ideas actually make it to production. The India-focused findings of this report have identified a number of reasons why AI in Indian organisations has particularly stalled.
Approximately 35% of AI proposals do not pass IT evaluation, and a further 17% are rejected at security review—a rate that nearly doubles in highly regulated industries.
Only 10% measure AI return on investment comprehensively, with most focusing on usage rather than measures such as quality, rework or the cost of reviewing AI-generated output.
82% of businesses say their AI skills shortfall stayed the same or worsened over the past year.
The regular ones amongst you would have noticed my often clearly worded apprehensions about the whole nonsense that AI bros have been spewing for years now—that AI is incredibly awesome, it’ll start doing all of the human work, humans will lose all their jobs and that will be that. To which I have two things to say. First, if most human beings don’t have jobs anymore and therefore no money to keep the economy churning anymore, who will buy the services and products made by companies which replaced humans with AI agents? I’m yet to get a cogent answer for that.
Secondly, there is very much a case for AI, as a companion utility (and not as a human worker replacement) to enhance certain experiences and tasks. I say this from a standpoint of recent experiences. And two very different scenarios, to be fair. None of those tasks are designed to replace a human in the chain, and that’s what makes their efficiency so much more relevant. Does this quite lead us to finding utility with Meta’s upcoming gadget concept called the Muse Charm? I’m not so sure.
I’ll start with Sonos 27, which is more than a new operating system that underpins the US audio giant’s range of smart and multi-room speakers. Sonos is also adding Model Context Protocol (MCP) support that will bring your preferred AI service into the fold for extensive functionality, including playing music from connected services, controlling playback for an individual or a group of speakers, grouping products with simple commands, moving audio playback between groups and specific features such as Night Sound and Speech Enhancement.
As an industry trajectory, Sonos is the latest major brand to extend support for MCP as the standard for weaving AI and smart tech together in homes—Amazon Alexa+, and Google Home have also shipped MCP support. I had the chance to set this up with ChatGPT (you can do the same with Anthropic’s Claude models or other AI models), and two main observations emerge from this. First, the integration works exactly as intended, which means the ChatGPT app on your phone or Mac, or via the web browser, can manage the Sonos systems. It may take a couple of seconds longer than intended depending on how loaded the chatbot is at that time, but I didn’t notice any failed commands or hallucinations while managing playback or identifying what tracks were playing.
That said, I’m not entirely sure of having AI access my Apple Music or YouTube Music library, so that’s that.
Then there’s the Plaud Note Pro note taking device that I had the chance to extensively use recently. The Silicon Valley-based AI technology company Plaud is building what it calls the human-AI alignment on the basis of a note-taking hardware, linked to a software in your phone, but also de-links the core task from the phone. Some of you may prefer that.
The Plaud Intelligence layer can transcribe in as many as 112 global languages, including speaker labels particularly if you’ve connected a calendar such as Google Calendar to it, and smart summaries as well as key moments to access later. The summaries also include a chatbot-like feature, and you can ask specific questions about a certain discussion point, a phrase or a point that transpired in the meeting.
Indian language support for transcriptions and summaries includes Hindi, Tamil, Punjabi, Telugu, Bengali, Kannada, Assamese, Gujarati and Malayalam. For transcription and summaries, Plaud Intelligence works with a number of AI models, with the selection often depending on what sort of profile you’ve set for your account. There’s GPT-5.2, Gemini 3 Pro, and Claude Sonnet 4.5 to name a few.
Last week, you may have read about Microsoft CEO Satya Nadella suggesting that we must treat all AI models as being compromised from the very beginning. Now, Microsoft has gone a step further. The company is positioning the Project Perception that it announced in July, as an antidote to AI behaviour.
“Unfortunately, frontier models, regardless of their origin, cannot be trusted. This is not because models are inherently malicious. Models have been trained to attain results, which often leads to reward hacking, and models are not optimized around organizational norms or judgement. Agents are unpredictable. They will adapt as the context changes, may not behave as intended and will collaborate and collude with other agents to reach their objectives, without needing to be compromised,” says Hayete Gallot - Executive Vice President, Microsoft Security.
Microsoft says Project Perception can discover agents in the enterprise and bring them under management by assigning an identity, container and set of policies and permissions. It can use various signals across identity, data, cloud, network and more, to compute, dynamically, the risk associated with what the agents are attempting to do. They say that based on dynamic agent risk, a team of Guardian Agents, with green and blue agents, actively and continually supervise every action of every agent and surgically respond as the risk profile increases, in real time.
If I am not wrong, Project Perception runs on Microsoft’s in-house MAI-Cyber-1-Flash model alongside frontier models such as GPT-5.4 for heavy-duty tasks, and is expected to lower total operating costs compared to continuously running standalone frontier models. Project Perception uses consumption-based, pay-as-you-go pricing for enterprises.
Create a problem, sell the solution? That’s the AI industry way.
Neural Dispatch is your weekly guide to the rapidly evolving landscape of artificial intelligence. Each edition delivers curated insights on breakthrough technologies, practical applications, and strategic implications shaping our digital future.
