. @ksr_swamy, Chief AI Officer at @Zscaler, joins FUSION'26! Security used to mean keeping AI out — now it's about bringing it in safely, his exact mandate at one of the world's top security companies. Sept 29, Palace Hotel, SF. Request your invite: fusion.atomicwork.com

Enterprise AI is just getting started, we are not even 1% done!

I disagree with this - Burnout is a function of energy management and not “how many obsessions” you have. @mattmochary has a great framework for an energy audit - what gives you energy, what takes away energy and what is energy neutral. You can get to “energy positive” in a myriad of ways - for some people it is taking out time to read a book or play a sport or spend time with fam and friends, for some people it is finding more sources of energy at work itself (and sometimes killing things that drain your energy). Sometimes trying to find a second obsession can be more energy draining than investing in your first obsession :)

Congrats Manohar, great run and excited to see what you do next 🎉👏

Be nice, it’s worth 💜

Human managers, meet your AI Coworkers. We're at Oktane showing what an AI Workforce delivers — more capacity without more headcount, 24/7 coverage, and a lot more than ticket deflection. The part nobody demos: identity. An AI Coworker has to act without you present and carry your specific authority on every call. Service accounts give you one; login gives you the other, neither gives you both. That's what stalls this in security review. Atomicwork is now live on the Okta Integration Network with Cross-App Access. Scoped, time-boxed, revocable from Okta alone! If you're at Oktane, come by Booth G1. I'd love to talk more 🙂

The AI race has two runners: China and the US. The EU is busy regulating. India is busy talking.

The loudest voices stoking fears about AI dangers have made tremendous headway in the past two weeks. AI technology has not taken some unexpected, dangerous turn, but the hype around it — propelled by what appears to be a well orchestrated PR campaign — has drummed up considerable fear. I worry that it represents a setback for our field. I have written frequently that fears of AI are overhyped. AI’s capabilities can be uncannily human-like and unpredictable, and it’s rational to worry when people who are directly involved express concerns. But I see the problems as a sign of the engineering work that ahead, rather than insurmountable barriers or the sky falling. AI technology continues to advance — which is a good thing! — but technical advances, poorly understood by the public, give those who seek to generate hype repeated opportunities to do so. First, I don’t see any step up in the risk of human extinction from AI compared to a few months ago. The theories about this remain the same fantastical, science fiction scenarios as a few months ago. The biggest change in AI risk is its cybersecurity capabilities — a topic which we should take seriously — but this, too, will not lead to the end of the world. The most notable recent event leading to increased fear was when an OpenAI team deployed an agent swarm that hacked into Hugging Face. Much of the popular press contained significant hype. For example, some publications reported that a swarm of 1,200 agents carried out the attack. While this was technically accurate, as I write this, I have about 1,300 processes running on my laptop. Yes, the ability to get large swarms of agents to work in parallel on a task is a significant technical advance, And, in computing, many processes run at the same time. So this shouldn’t be seen as some magical capability. Additionally, OpenAI’s buggy sandboxing and monitoring processes were key to enabling this incident. Fixing these bugs and putting in place improved monitoring would be appropriate fixes, not pausing AI. There are many well known ways to attack software systems. The main advantage of AI agents is that they are relentless. They will tirelessly try many tactics — and have the patience to chain vulnerabilities together — that previously would have taken an infeasible amount of human effort. But in the long term, I believe the advantage will lie with defenders (because they have more information with which to identify bugs, which they can fix), but the cyber-threat landscape has changed significantly. There are still bottlenecks to identifying and exploiting a vulnerability. AI agents still have to try a lot of things to see what works, and taking these actions takes time and might be detected by defenders. This is why, even though it is now easy to obtain versions of leading open weight models that have had their guardrails removed or weakened, so they will not refuse to try to execute cyber attacks, the world has not ended. I am also concerned about the anthropomorphization of AI in a lot of reporting, where LLMs and agents are unnecessarily treated as if they were people. If I wield a hammer, miss a nail, and accidentally dent the wall, it’s not the fault of the hammer. The problem lies in how I used the hammer. Similarly, if I prompt an agent and it hacks into someone else’s system, the responsibility lies with me, not the agent. Of course, we want to build systems that are as safe and predictable as possible. (For example, an unsafe hammer would be one whose head randomly flies off under normal use.) Today’s agentic systems are not predictable, but I see no reason why, by applying sound engineering practices, we won’t be able to make them extremely safe to use. One new element in the forecasts of AI-enabled doom is AI companies disclaiming responsibility for their own products. “I didn’t do it; my out-of-control agent did!” There’s a balance to be struck between the responsibility of the tool maker and the tool user, but when something goes wrong, let’s hold the people building and/or using the hammer responsible, rather than the hammer. (By the way, if you’re worried about AI bioweapon risk, David Bellamy has a great post on why this, too, is overhyped. Briefly, the bottleneck in building a bioweapon is not intelligence, but lab work and manufacturing.) Pausing AI progress will create much more harm than benefit. First, our adversaries will certainly not slow down. Second, engineering requires discovering problems empirically so we can fix them. If we pause AI by a decade, we will also delay finding and implementing safety engineering fixes by about the same duration. Of course, the incentive to stoke fears — for regulatory capture, to garner attention, or to make one’s technology seem more powerful — remains the same as before. Disclaiming responsibility is a new one. Taking a hard technical look at the actual risks however, I see little factual basis for the degree of fear that’s been stoked up. We still have hard research and engineering work ahead to improve AI safety, but the beneficial applications continue to vastly outweigh the risks, and we should keep building. [Original text (with links): deeplearning.ai/the-batch/is… ]

Much needed to unleash agentic intelligence inside enterprises. Congrats on the launch @okta team 💜🎉🚀

Marcus Aurelius nailed it. The moment you are disturbed by insult or pleased by praise, you are still a slave.

AI VC Thought Leadership. Create the hype. Fund the crap. Inflate the value. And then tweet that 99% of AI startups will die. My friend, you set the price!

The first gift is life. The second gift is good health. The third gift is to love and be loved. The fourth gift is peace of mind. The fifth gift is having the wisdom to know what to appreciate most. There are no other gifts that matter as much.

Jev's popularity is Jevons paradox playing out in AI ecosystem. Jev model is an interesting signal for where agentic AI architecture is heading for both consumer and enterprise use cases. A surprising amount of work inside AI agents and solutions like @atomicworkhq AI Coworkers is low entropy tasks like classify, route, score, pick a tool and check a condition. Yet we often use autoregressive LLMs built to generate language for tasks that only need a decision. Jev's architecture flips that model. For the many workloads, one-pass typed decisions can deliver incredibly lower cost and latency, with some multi-step AI workflows like browser-use showing 100x+ improvements. However, the bigger insight is Jevons paradox. Make intelligence 100x cheaper and faster, and we probably won't use 100x less of it. We'll use it 1,000x more. The future of AI agents and AI workforce will not use use just frontier LLM models for doing everything. Cheap System 1 models like Jev for low-entropy decisions. Frontier LLM models for high-entropy reasoning. Incredible launch from @typesafeai and congrats @CompleteSkeptic for reigniting AI build energy on X.💜🔥

Heard @alighodsi speak twice at the Khosla CEO Summit under Chatham House rules. He’s an incredible storyteller, but what stood out to me is how deeply he understands culture, technology, GTM and strategy. The story of how he built, scaled and reinvented Databricks is going to be one hell of a case study for generations.

Meta’s Muse Eats into Instinct’s Early Lead The personal AI agent war is upon us. For weeks now, tech insiders have been drooling over Instinct, a months-old startup that works as a chat box on WhatsApp or iMessage and books tickets, finds apartments, sends emails, and cancels subscriptions, all for free so far. But even as Benchmark-backed Instinct reportedly seeks a $10 billion valuation, Meta is nipping at its heels. Meta’s Muse, launched last week, connects to each app on your phone and can run the mundane parts of life with its own browser. The browser is visible on your screen, while Instinct runs everything on its backend, with only a chat box visible to you. Instinct has the happy problem — but problem nevertheless — of slowing answers due to capacity constraints, as we’ve experienced along with many others. Meta is aiming to capitalize. Vijay Rayapati, who runs Atomicwork, an AI rival to ServiceNow, said Muse feels faster and he is using it much more than Instinct, a sentiment echoed by many on X. Muse also works on “isolated compute,” Rayapati said, making it faster and more reliable for privacy purposes, compared to Instinct which distributes compute across users. Both companies also have to compete with Town, backed by Andreessen Horowitz and Forerunner Ventures. It is obviously far too early to call any winners, but Meta’s entry, and early success, shows the power that Big Tech incumbents still enjoy. Instinct’s 23-year-old CEO Noah Shinn, who seems to communicate with the world mainly via X, is rolling out new features at breakneck speed, though he doesn’t acknowledge the competition. In recent days, Instinct can now make phone calls, and talk to your friends’ Instinct agents to coordinate plans, taking it deeper into people’s lives. More at @NewcomerMedia

When @Benioff likes your satire about Dreamforce, you know he can take a joke. Legend for a reason. 💙🫶 Thanks Marc for playing big daddy and saving SaaS from the AIpocalypse with #DF26.

Harness Tax is real. Enterprise AI needs to move beyond cost per token to cost per verified outcome. Cost per Verified Outcome = f(Model, Reasoning Effort, Harness, Context, Tools, Workload) In 2024, I wrote about the CAP constraint in AI apps: Cost, Accuracy, Performance. A year of building AI Coworkers at @atomicworkhq has made those tradeoffs even clearer. What we’ve learned in production: atomicwork.com/blog/harness-…

Pretty accurate

Four years of Atomicwork. Feels like yesterday. In September 2022, we made a bet most people thought was early. We believed that the moment AI could actually do the work, the queue itself would become the problem. So we started the AI-native ITSM category before anyone was calling it that. Four years later, the thesis is clearer than ever. We are not building another AI Agent and we are not stitching AI onto code scripted workflows. We are building the AI Workforce for Enterprise IT and believe that AI will become a new workforce operating model for enterprises: Humans + AI Coworkers. AI Coworkers that take the work, do it end to end and hand it back by operating at job role level not workflow level. An IT team that hires them into job roles and governs them like HR does for people. That's the difference between AI as a feature and AI as a workforce. Grateful to the team that came early on conviction, the customers who trusted us and the partners who backed this before it was obvious. It was so much fun this time as we had customers and families join our celebration. Year five. Let's go. IT's our time! 💜🚀