¶ Introduction to AI at CES 2026
Hey everyone. I'm Andy, and this is the AI breakdown, CES 2026 wrapped up in Las Vegas earlier this month, and I think it quietly marked a shift. AI has moved on from being the shiny add-on. You stick on a product to get a headline and started behaving like infrastructure, like wifi, like cloud, like electricity. It's just assumed, and you could feel it everywhere. The keynotes weren't really about novelty anymore. They were about integration.
AMD's, CEO summed up the mood with a simple line saying AI is everywhere and for everyone. So in this episode, I'm going to walk you through five big AI themes.
CES made impossible to ignore, and I'll keep it grounded in what was actually shown and announced. 11 00:00:50,41.126688514 --> 00:00:51,361.126688514 All right, let's get into it. 12 00:00:51,751.126688514 --> 00:01:01,381.126688514 When people say AI is everywhere, it can sound like marketing fluff at CES 2026, though it was unusually literal and the most important place it showed up. 13 00:01:01,426.126688514 --> 00:01:04,216.126688514 It wasn't in smart fridges or talking mirrors. 14 00:01:04,816.126688514 --> 00:01:06,196.126688514 It was in the workhorse device. 15 00:01:06,226.126688514 --> 00:01:08,656.126688514 Most businesses buy in bulk, the humble laptop. 16 00:01:09,526.126688514 --> 00:01:11,476.126688514 The anchor story for this theme is simple. 17 00:01:11,746.126688514 --> 00:01:17,396.126688514 Your next work PC is going to have an AI copilot under the hood whether you ask for one or not. 18 00:01:18,451.126688514 --> 00:01:23,641.126688514 Now before your eyes glaze over, let me quickly translate what the chip companies were all talking about. 19 00:01:24,91.126688514 --> 00:01:29,491.126688514 They kept throwing around this term tops that stands for trillions of operations per second. 20 00:01:30,1.126688514 --> 00:01:36,991.126688514 Basically, a fancy way of seeing how much AI thinking your laptop can do on its own without sending everything off the cloud. 21 00:01:37,471.126688514 --> 00:01:45,601.12668851 EMD used CES to launch its new Ryzen AI 400 series chips boasting up to 60 tops on what's called the NPU. 22 00:01:46,21.12668851 --> 00:01:51,151.12668851 Or more specifically, the neural processing unit, a little dedicated AI engine. 23 00:01:51,151.12668851 --> 00:01:59,431.12668851 Inside the processor, Intel came back with its core Ultra Panther Lake Chips, claiming a whopping 180 tops across the whole system. 24 00:02:00,211.12668851 --> 00:02:08,971.12668851 And what Intel is really seeing there is that your laptop isn't just going to run spreadsheets, it's going to run an assistant locally, privately, even offline. 25 00:02:10,471.12668851 --> 00:02:25,51.12668851 Qualcomm meanwhile is pushing AI into the mid-range with its Snapdragon X two plus at 80 tops, and it's making the battery life pitch two saying it cuts AI power use by roughly 43%, so assistance can run all day. 26 00:02:25,681.12668851 --> 00:02:29,851.12668851 So yes, the numbers are big, but the real message from CES is even bigger. 27 00:02:30,466.12668851 --> 00:02:35,206.12668851 Stop thinking of on device AI as acute webcam filter, or a gimmick. 28 00:02:35,776.12668851 --> 00:02:39,496.12668851 This is serious capability being baked into everyday business computers. 29 00:02:39,826.12668851 --> 00:02:45,16.12668851 The same ones sitting on desks in EHR, finance, safety teams and operations. 30 00:02:45,676.12668851 --> 00:02:47,806.12668851 So AI isn't arriving as an app anymore. 31 00:02:47,986.12668851 --> 00:02:49,726.12668851 It's arriving as part of the hardware. 32 00:02:50,146.12668851 --> 00:02:57,796.12668851 Microsoft leaned into this with its new copilot plus PC framing, basically positioning these as the smartest and most secure. 33 00:02:58,111.12668851 --> 00:03:06,181.12668851 Windows laptops yet because they come with dedicated AI hardware built in and the manufacturers, they're already lining up to ship them. 34 00:03:06,601.12668851 --> 00:03:16,561.12668851 Lenovo, for example, introduced ThinkPad X one aura additions as part of the A IPC wave, and there were even concept devices pitched around ambient ai. 35 00:03:17,41.12668851 --> 00:03:19,981.12668851 The idea that your device proactively adapts to you. 36 00:03:20,341.12668851 --> 00:03:23,191.12668851 Now, here's the business angle, and it's bigger than people think. 37 00:03:23,491.12668851 --> 00:03:28,411.12668851 Procurement and it refresh cycles are about to become AI strategy decisions. 38 00:03:28,741.12668851 --> 00:03:36,181.12668851 For years, the laptop refresh conversation was basically price performance, battery life warranty may be security certifications. 39 00:03:36,181.12668851 --> 00:03:44,251.12668851 Now you're adding a new line item on device AI capability because if you buy a fleet of machines with strong NPUs. 40 00:03:44,971.12668851 --> 00:03:51,181.12668851 You are implicitly deciding that parts of your organization's AI workload can and should run locally. 41 00:03:51,571.12668851 --> 00:03:52,861.12668851 That has knock on effects. 42 00:03:53,221.12668851 --> 00:04:00,631.12668851 It can change your cloud spend, it can change your security posture, and it can change what AI access looks like for employees. 43 00:04:01,171.12668851 --> 00:04:01,981.12668851 Think about it. 44 00:04:02,311.12668851 --> 00:04:12,661.12668851 If a laptop can do local transcription, summarization, translation, document search, maybe even run a decent sized assistant offline, then the employee experience shifts. 45 00:04:13,21.12668851 --> 00:04:17,641.12668851 The sales person on a train with patchy wifi can still have an AI helper. 46 00:04:18,181.12668851 --> 00:04:22,471.12668851 The finance team can analyze sensitive documents without pushing them to an external server. 47 00:04:22,921.12668851 --> 00:04:26,161.12668851 Your legal team can summarize contracts without a cloud round trip. 48 00:04:26,821.12668851 --> 00:04:31,496.12668851 CES also hinted at this becoming a developer opportunity, not just an end user feature. 49 00:04:32,256.12668851 --> 00:04:39,811.12668851 EMD even introduced a mini PC Ryzen AI halo, positioned as a developer kit for High Performance Edge ai. 50 00:04:40,186.12668851 --> 00:04:46,966.12668851 Basically seeing, here's a box you can use to prototype generative AI apps that run at the edge, not the data center. 51 00:04:47,566.12668851 --> 00:04:56,776.12668851 So if you are a software company or even an internal IT team building custom tools, you can start to treat the NPU as a new deployment target. 52 00:04:57,286.12668851 --> 00:05:03,496.12668851 Just like we once learned to target the GPU for graphics, now we target the NPU for local inference. 53 00:05:03,901.12668851 --> 00:05:06,181.12668851 But infrastructure cuts both ways. 54 00:05:06,391.12668851 --> 00:05:08,221.12668851 First, vendor lock-in is real. 55 00:05:08,701.12668851 --> 00:05:13,651.12668851 When AI is baked into hardware, you're not just buying a laptop, you're buying an ecosystem. 56 00:05:14,341.12668851 --> 00:05:17,851.12668851 Different chip makers have their own stacks and optimization paths. 57 00:05:18,301.12668851 --> 00:05:27,571.12668851 If you build internal tools that only run well on one vendor's NPU, you've created a subtle dependency that will show up in your next procurement negotiation. 58 00:05:28,156.12668851 --> 00:05:31,96.12668851 Second management overhead is about to get weird. 59 00:05:31,336.12668851 --> 00:05:36,676.12668851 We're used to managing devices and apps, but managing models across thousands of devices. 60 00:05:36,706.12668851 --> 00:05:37,816.12668851 That's new territory. 61 00:05:38,416.12668851 --> 00:05:40,666.12668851 CES's own trend analysis flag. 62 00:05:40,666.12668851 --> 00:05:44,116.12668851 This where it complexity becomes the hidden cost. 63 00:05:44,536.12668851 --> 00:06:02,581.12668851 You've got questions like, how do we update models safely? How do we ensure consistency? How do we store models securely on device? What happens when one department is running an older model that behaves slightly differently from another? There's also the reality that the hardware is arriving faster than the software is maturing. 64 00:06:03,241.12668851 --> 00:06:08,881.12668851 A few years ago when a I PCs first hit the market, we were all asking the same question. 65 00:06:09,181.12668851 --> 00:06:23,761.12668851 What actually uses the NPU? CES 2026 felt like the industry is finally answering that, but it still requires vendors to optimize properly and for businesses to be thoughtful about what they deploy locally versus centrally. 66 00:06:24,241.12668851 --> 00:06:28,921.12668851 My opinion is to treat AI PCs the way you treated cloud 10 years ago. 67 00:06:29,401.12668851 --> 00:06:32,581.12668851 Don't buy random instances and hope it sorts itself out. 68 00:06:32,826.12668851 --> 00:06:33,646.12668851 Create a standard. 69 00:06:34,276.12668851 --> 00:06:41,356.12668851 Decide what workloads you want on device, what you want in the cloud, and what you absolutely don't want happening on unmanaged endpoints. 70 00:06:41,716.12668851 --> 00:06:50,476.12668851 And if you are in procurement, a laptop refresh might now be the biggest AI decision your company makes this year, not because the laptop itself is magical. 71 00:06:51,286.12668851 --> 00:06:57,856.12668851 But because it determines where your AI runs, how your data moves, and how much control you actually have. 72 00:06:58,186.12668851 --> 00:07:08,416.12668851 So yes, AI becoming baseline infrastructure is exciting, but the real maturity move is recognizing that AI inside isn't a feature, it's an operational commitment. 73 00:07:10,87.94487033 --> 00:07:20,557.94487033 The second CAS theme is the one that makes the infrastructure story feel real edge, AI and privacy, or as I like to call it, the no cloud required wave. 74 00:07:21,307.94487033 --> 00:07:24,997.94487033 At CES 2026, the shift wasn't theoretical. 75 00:07:25,387.94487033 --> 00:07:31,627.94487033 It showed up in demos where the whole point was that AI could function even if the internet fell over. 76 00:07:32,407.94487033 --> 00:07:34,657.94487033 Let's run through three quick anchor stories. 77 00:07:34,987.94487033 --> 00:07:37,447.94487033 First, deepfake detection on device. 78 00:07:37,897.94487033 --> 00:07:46,177.94487033 Jen, the company behind Norton, partnered with Intel to show a real time d fake detection prototype, running locally on a pc. 79 00:07:47,62.94487033 --> 00:07:54,802.94487033 The key detail here is that it analyzed video and audio simultaneously on device designed to catch manipulation as it happens. 80 00:07:55,492.94487033 --> 00:08:01,282.94487033 Jen said that research found scammers increasingly embed dfas inside long form videos. 81 00:08:01,582.94487033 --> 00:08:07,582.94487033 Think lengthy YouTube streams or Facebook videos where the scam slowly builds toast before the hook. 82 00:08:08,152.94487033 --> 00:08:14,92.94487033 So the detection tool isn't just looking at the first few seconds, it's monitoring the content while it plays. 83 00:08:14,662.94487033 --> 00:08:16,222.94487033 That's a proper shift in posture. 84 00:08:16,762.94487033 --> 00:08:25,522.94487033 Instead of telling employees, be careful out there and doing awareness training, you start giving them tooling that can flag suspicious, medium in real time. 85 00:08:26,2.94487033 --> 00:08:29,392.94487033 Second on device KYC and deepfake checks. 86 00:08:29,782.94487033 --> 00:08:39,322.94487033 A starter called true sources showing in Eureka Park, demoed on device identity verification that does liveness checks and deepfake detection locally. 87 00:08:39,532.94487033 --> 00:08:41,992.94487033 So your personal data never leaves your phone. 88 00:08:42,382.94487033 --> 00:08:46,642.94487033 That's a privacy story, yes, but it's also a product differentiator story. 89 00:08:46,972.94487033 --> 00:08:52,792.94487033 In regulated industries, not having to ship sensitive identity data to a server is a big deal. 90 00:08:53,452.94487033 --> 00:08:56,242.94487033 Third Caterpillar's, Kati assistant. 91 00:08:56,647.94487033 --> 00:09:01,717.94487033 This is where Edge AI stops being about convenience and starts being about safety. 92 00:09:02,497.94487033 --> 00:09:22,597.94487033 Caterpillar's CEO Joe Creed used a CES keynote to launch the Cat AI assistant, a voice activated helper for equipment operators in a live demo and operator used Hey Cat, to set a safety limit and the excavator automatically kept the boom from hitting a 13 foot overhead power line. 93 00:09:23,257.94487033 --> 00:09:32,557.94487033 Caterpillar said the offboard version would roll out via its apps and websites by Q1 2026 with the in-cab version and final validation. 94 00:09:33,67.94487033 --> 00:09:41,347.94487033 And the reason this fits the Edge AI theme is that the in machine assistant, as shown at CES, was designed to run locally on the equipment. 95 00:09:41,797.94487033 --> 00:09:52,117.94487033 Caterpillar demoed it running on an Nvidia Jetson edge computer with a local 4 billion parameter language model with no cloud link required for real-time guidance. 96 00:09:52,822.94487033 --> 00:10:02,572.94487033 This matters because Edge AI solves three very unsexy, very real problems like latency, offline reliability, and data control. 97 00:10:03,232.94487033 --> 00:10:13,252.94487033 If you're doing something time sensitive like an operator moving a six ton excavator arm near a power line, you don't want your safety boundary check to wait on a network round trip. 98 00:10:13,972.94487033 --> 00:10:18,742.94487033 And thinking of remote sites and field operations that often have patchy connectivity. 99 00:10:19,252.94487033 --> 00:10:22,642.94487033 Caterpillar explicitly has to design for this type of reality. 100 00:10:23,212.94487033 --> 00:10:29,602.94487033 Edge AI means the system still works in mud, dust, and chaos, not just in a well-connected office. 101 00:10:29,992.94487033 --> 00:10:35,572.94487033 And when it comes to data control for a lot of businesses, the barrier to AI adoption is simple. 102 00:10:35,902.94487033 --> 00:10:43,852.94487033 They don't want to send sensitive data to someone else's servers on device processing changes that it's not perfect, but it's a big step. 103 00:10:44,392.94487033 --> 00:10:48,652.94487033 Now, there is a reality check here on device models are smaller. 104 00:10:49,12.94487033 --> 00:10:50,32.94487033 That's not a complaint. 105 00:10:50,32.94487033 --> 00:10:50,872.94487033 It's physics. 106 00:10:51,232.94487033 --> 00:10:59,332.94487033 You're not running a massive cloud model on a laptop NPU or a Jetson in an excavator and getting the same performance across every task. 107 00:10:59,812.94487033 --> 00:11:07,372.94487033 These edge deployments are about choosing the right model for the right job, and then you hit the scaling headache, like updates and maintenance. 108 00:11:07,822.94487033 --> 00:11:17,687.94487033 If you've got AI models sitting on thousands of laptops, phones, vehicles, or machines, you need a way to push updates, manage versions, and make sure the model you validated last month. 109 00:11:18,367.94487033 --> 00:11:20,257.94487033 It's still the model running in the field. 110 00:11:20,887.94487033 --> 00:11:30,517.94487033 C S's own analysis called this Out Edge AI becomes a new layer of infrastructure to manage like mobile device management, but now it includes AI models. 111 00:11:31,57.94487033 --> 00:11:35,797.94487033 So my takeaway is not move everything to the edge, it's put AI where the work is. 112 00:11:37,255.90515824 --> 00:11:40,405.90515824 Theme three is the one I hope kills off the most tedious phrase. 113 00:11:40,405.90515824 --> 00:11:42,685.90515824 In tech, we're exploring ai. 114 00:11:43,540.90515824 --> 00:11:48,160.90515824 At CES 2026, the tone shifted toward return on investment. 115 00:11:48,160.90515824 --> 00:11:53,470.90515824 First, AI decision intelligence and digital twins, where the bar's very simple. 116 00:11:53,680.90515824 --> 00:11:54,850.90515824 Show me the numbers. 117 00:11:55,540.90515824 --> 00:11:57,640.90515824 Siemens delivered the cleanest example. 118 00:11:57,910.90515824 --> 00:12:07,995.90515824 They launched a digital twin composer platform as part of their accelerator suite built to create full 3D physics, accurate digital twins of factories or supply chains. 119 00:12:08,710.90515824 --> 00:12:14,530.90515824 You feed them real time data and then use AI agents to test optimizations virtually. 120 00:12:15,70.90515824 --> 00:12:19,930.90515824 And they didn't just talk about it in theory, they highlighted PepsiCo as a real world user. 121 00:12:20,440.90515824 --> 00:12:25,870.90515824 PepsiCo converted several US plants into digital twins and ran AI simulations. 122 00:12:26,410.90515824 --> 00:12:34,780.90515824 Siemens said the results were big, about a 20% throughput boost around a 10 to 15% reduction in CapEx on initial deployments. 123 00:12:35,140.90515824 --> 00:12:35,530.90515824 And. 124 00:12:35,890.90515824 --> 00:12:42,550.90515824 This is my favorite line, about 90% of potential issues found upfront before making physical changes. 125 00:12:43,150.90515824 --> 00:12:45,400.90515824 That last stat is the quiet killer feature. 126 00:12:45,760.90515824 --> 00:12:56,260.90515824 If you can spot most of the problems before you move a machine, change a line, or invest in new equipment, you're not just saving money, you're saving time, stress, and reputation. 127 00:12:56,800.90515824 --> 00:13:03,220.90515824 Siemens also announced nine new industrial AI co-pilots across design, engineering, and operations. 128 00:13:03,595.90515824 --> 00:13:11,215.90515824 Tools that do the boring but huge work, like automatically checking product data for errors or handling compliance documentation. 129 00:13:11,725.90515824 --> 00:13:16,555.90515824 The business angle here is that AI projects are no longer competing against doing nothing. 130 00:13:16,885.90515824 --> 00:13:19,705.90515824 They're competing against hard r oi opportunities. 131 00:13:20,125.90515824 --> 00:13:29,665.90515824 If a digital twin can realistically lift throughput by 20%, then your cool chatbot pilot with no clear KPI starts to look a bit unserious. 132 00:13:30,205.90515824 --> 00:13:32,35.90515824 And here's what CES made clear. 133 00:13:32,440.90515824 --> 00:13:34,360.90515824 The new bar is metrics or burst. 134 00:13:34,570.90515824 --> 00:13:37,600.90515824 You need to find KPIs and you need change management. 135 00:13:37,960.90515824 --> 00:13:39,760.90515824 But again, reality check. 136 00:13:40,210.90515824 --> 00:13:43,450.90515824 Those PepsiCo numbers don't come from model magic alone. 137 00:13:44,140.90515824 --> 00:13:47,800.90515824 CES commentary around these winds kept circling back to the same thing. 138 00:13:48,160.90515824 --> 00:13:53,650.90515824 Clean well structured data is the fuel, and the other ingredient is process redesign. 139 00:13:54,130.90515824 --> 00:14:00,130.90515824 If your data is messy, siloed, and unreliable, your digital twin is going to be a beautiful 3D lie. 140 00:14:00,565.90515824 --> 00:14:07,885.90515824 And even if your simulation finds a better schedule or layout, you only get ROI if your organization can actually act on it. 141 00:14:08,335.90515824 --> 00:14:12,565.90515824 That means policies, incentives, and frontline workflows have to change. 142 00:14:12,715.90515824 --> 00:14:29,485.90515824 So if you're listening and thinking, right, we need a digital twin, my advice is start smaller than you want to pick a process with a measurable pain point, downtime, scrap rate throughput, CapEx overruns, and build a capability to simulate and test changes. 143 00:14:30,115.90515824 --> 00:14:37,555.90515824 Make sure you can measure outcomes in the real world and invest in the plumbing data quality, data governance and integration. 144 00:14:39,132.94090288 --> 00:14:48,252.94090288 Theme four is physical ai, and yes, it sounds like something a marketing team made up, but CES 2026 showed why the phrase is sticking. 145 00:14:48,822.94090288 --> 00:14:55,32.94090288 This is AI leaving the screen and turning up in machines that can sense, decide and act in the real world. 146 00:14:55,482.94090288 --> 00:14:57,12.94090288 The contrast story was clear. 147 00:14:57,582.94090288 --> 00:15:05,922.94090288 On one end you've got Humanoids Boston Dynamics showed its Atlas Humanoid as fully electric and production ready for industrial use. 148 00:15:06,282.94090288 --> 00:15:09,612.94090288 Hyundai I says it plans to deploy Atlas at a Georgia plant. 149 00:15:09,912.94090288 --> 00:15:16,182.94090288 By 2028 for pot sequence and tasks and the framing was human robot collaboration. 150 00:15:16,542.94090288 --> 00:15:25,752.94090288 Atlas can lift 110 pounds, has 56 degrees of freedom and uses Google deepminds, Gemini AI for advanced planning and problem solving. 151 00:15:26,172.94090288 --> 00:15:27,582.94090288 That's genuinely impressive. 152 00:15:27,882.94090288 --> 00:15:32,832.94090288 On the other end, you've got heavy machinery core pilots, which are nearer term for a lot of industries. 153 00:15:33,342.94090288 --> 00:15:38,382.94090288 Caterpillar's cat AI assistant demo that I talked about earlier is the perfect example. 154 00:15:38,712.94090288 --> 00:15:47,952.94090288 A voice assistant embedded into a six ton excavator setting safety boundaries and pulling troubleshooting, guidance and manuals on request designed to operate locally. 155 00:15:48,702.94090288 --> 00:15:58,152.94090288 If you run construction, mining, agriculture, logistics, or anything with expensive equipment and a skills gap, the copilot model makes immediate sense. 156 00:15:58,512.94090288 --> 00:16:01,302.94090288 It doesn't require a full humanoid robot revolution. 157 00:16:01,647.94090288 --> 00:16:06,387.94090288 It upgrades the tools you already have and caterpillar's broader context matters. 158 00:16:06,897.94090288 --> 00:16:12,522.94090288 Caterpillar's Chief Digital Officer talked about combining AI with its fleet of 1.5 159 00:16:12,522.94090288 --> 00:16:19,257.94090288 million connected machines, feeding 16 petabytes of operational data into its Helios platform. 160 00:16:19,677.94090288 --> 00:16:25,287.94090288 That's the kind of proprietary data mode that makes domain specific physical AI plausible. 161 00:16:25,977.94090288 --> 00:16:27,207.94090288 Now, the business angle. 162 00:16:27,627.94090288 --> 00:16:30,297.94090288 Physical AI doesn't just change efficiency. 163 00:16:30,627.94090288 --> 00:16:34,557.94090288 It changes labor safety insurance and operations design. 164 00:16:35,187.94090288 --> 00:16:42,327.9409029 If you introduce an AI copilot that prevents power line strikes, that's not just a productivity win, it's downtime avoided. 165 00:16:42,567.9409029 --> 00:16:48,567.9409029 It's fewer incidents, it's potentially fewer claims, and it's a different training curve for new operators. 166 00:16:48,927.9409029 --> 00:16:55,557.9409029 If humanoids like Atlas really do move into factories over the next few years, that changes how you design work cells. 167 00:16:55,872.9409029 --> 00:16:57,702.9409029 And how you allocate human labor. 168 00:16:58,362.9409029 --> 00:17:02,202.9409029 Hyundai's own timeline hints at phased adoption for a reason. 169 00:17:02,832.9409029 --> 00:17:07,212.9409029 The reality check is that the physical world is less forgiving than software. 170 00:17:07,752.9409029 --> 00:17:08,982.9409029 CES even had. 171 00:17:09,12.9409029 --> 00:17:16,332.9409029 LG showing a home robot clawed that can fold laundry and load a dishwasher and LG openly admitted it's slow. 172 00:17:16,872.9409029 --> 00:17:18,822.9409029 That's more OFX paradox in action. 173 00:17:19,122.9409029 --> 00:17:24,132.9409029 Physical skills remain hard, so my recommendation is pretty boring, but useful. 174 00:17:24,567.9409029 --> 00:17:31,857.9409029 If you're an operations leader, focus on physical AI that reduces errors, improves safety, and augments skilled work. 175 00:17:31,857.9409029 --> 00:17:40,527.9409029 First, copilots guided workflows, vision safety systems, and track humanoids as a medium term bit rather than tomorrow's staffing plan. 176 00:17:41,37.9409029 --> 00:17:42,687.9409029 The hype is robots are here. 177 00:17:43,107.9409029 --> 00:17:49,647.9409029 The practical truth is assistive machines are here too, and they're going to quietly change operations before the humanoids do. 178 00:17:51,225.9011908 --> 00:17:54,75.9011908 Theme five is AI across the customer journey. 179 00:17:54,420.9011908 --> 00:17:59,940.9011908 And CES 2026 made it clear this is moving beyond the dumb chatbot era. 180 00:18:00,660.9011908 --> 00:18:22,200.9011908 Two examples stood out in the research, first contact centers, CES, demos and commentary highlighted context aware bots, AI agent co-pilots for human reps, and omnichannel continuity, meaning you start in chat, hop to phone and nobody makes you repeat your life story because the context follows the broader claim floating around CES coverage. 181 00:18:22,560.9011908 --> 00:18:35,130.9011908 Is that advanced AI and contact centers can cut customer service costs by around 30% while speeding up resolution times, and that agent assist tools have been linked to big jumps in customer satisfaction in some reports. 182 00:18:35,610.9011908 --> 00:18:37,410.9011908 Second, marketing automation. 183 00:18:37,920.9011908 --> 00:18:46,470.9011908 Reddit used CES to unveil max campaigns and AI powered ad product that automates targeting, creative and optimization. 184 00:18:47,130.9011908 --> 00:18:50,790.9011908 Reddit said it had been alpha testing with more than 600 advertisers. 185 00:18:51,165.9011908 --> 00:19:00,285.9011908 And claimed early testers saw an average 27% higher conversion rate and 17% lower cost per action compared to their normal campaigns. 186 00:19:00,915.9011908 --> 00:19:05,955.9011908 It even generates Reddit ready headlines using trending subreddit, lingo, and auto crops. 187 00:19:05,955.9011908 --> 00:19:06,525.9011908 Creative. 188 00:19:07,155.9011908 --> 00:19:09,225.9011908 The business impact is pretty straightforward. 189 00:19:09,675.9011908 --> 00:19:14,595.9011908 AI can scale customer interactions and marketing decisions faster than teams can. 190 00:19:15,90.9011908 --> 00:19:18,660.9011908 It can run the funnel and support journey as an always on system. 191 00:19:18,990.9011908 --> 00:19:20,220.9011908 But the risks matter here. 192 00:19:20,610.9011908 --> 00:19:29,880.9011908 If you ever rely on AI in customer facing moments, you can end up with uniformity, every brand sounding the same, because they're all pulling from similar model patterns. 193 00:19:30,360.9011908 --> 00:19:37,620.9011908 Or you can misalign with your brand voice, and if your AI gets something wrong in a support context, you're not just wrong. 194 00:19:37,770.9011908 --> 00:19:38,850.9011908 You're wrong at scale. 195 00:19:39,465.9011908 --> 00:19:40,695.9011908 Treat customer journey. 196 00:19:40,755.9011908 --> 00:19:47,325.9011908 Ai, like a product, not a bolt-on, measure it, supervise it, and be clear about where humans step in. 197 00:19:47,805.9011908 --> 00:19:49,965.9011908 Because the goal isn't more automation. 198 00:19:50,265.9011908 --> 00:19:53,535.9011908 The goal is better customer experience with less waste. 199 00:19:54,830.5830832 --> 00:19:56,450.5830832 So let's pull all of this together. 200 00:19:56,810.5830832 --> 00:20:03,50.5830832 The real story is that AI is becoming an integral business component, not a nice to have enhancement. 201 00:20:03,470.5830832 --> 00:20:09,50.5830832 It's getting baked into the devices your people use, the operations that run your factories and fleets. 202 00:20:09,440.5830832 --> 00:20:11,360.5830832 And the systems that touch customers. 203 00:20:11,690.5830832 --> 00:20:14,930.5830832 But CES also made another point quietly and repeatedly. 204 00:20:15,500.5830832 --> 00:20:18,530.5830832 The limiting factor isn't clever models, it's trust. 205 00:20:19,220.5830832 --> 00:20:23,720.5830832 We saw that in the focus on on device deepfake detection with Gene and Intel. 206 00:20:24,140.5830832 --> 00:20:30,680.5830832 We saw it in true sources pitch, that verification can happen without shipping personal data off the phone. 207 00:20:31,100.5830832 --> 00:20:33,80.5830832 And we saw it in industrial ai. 208 00:20:33,440.5830832 --> 00:20:36,650.5830832 Where a simulation is only useful if you trust the data, feeding it. 209 00:20:37,100.5830832 --> 00:20:41,60.5830832 So here's the practical playbook I'm taking out of CES 2026. 210 00:20:41,510.5830832 --> 00:20:43,550.5830832 First move from pilots to platforms. 211 00:20:43,910.5830832 --> 00:20:45,290.5830832 The pilot phase is over. 212 00:20:45,680.5830832 --> 00:20:51,620.5830832 If you're still doing one-off AI experiments, you're going to waste time, build reusable capability. 213 00:20:52,70.5830832 --> 00:20:55,220.5830832 Data pipelines, model evaluation, governance. 214 00:20:55,520.5830832 --> 00:20:58,670.5830832 So every new AI use case gets cheaper and safer to ship. 215 00:20:59,180.5830832 --> 00:21:02,420.5830832 Second, treat edge and privacy as differentiators. 216 00:21:02,945.5830832 --> 00:21:05,525.5830832 On device AI isn't just a constraint. 217 00:21:05,825.5830832 --> 00:21:10,565.5830832 It's a way to offer faster, safer experiences and to operate reliably in the field. 218 00:21:11,15.5830832 --> 00:21:13,535.5830832 Third, make ROI your North Star. 219 00:21:14,45.5830832 --> 00:21:16,475.5830832 Copy what Siemens and PepsiCo showcased. 220 00:21:16,835.5830832 --> 00:21:22,805.5830832 Pick problems with measurable pain and insist on at least two metrics per rollout so you don't fool yourself. 221 00:21:23,345.5830832 --> 00:21:27,5.5830832 Fourth, invest in data and frontline workflow redesign. 222 00:21:27,560.5830832 --> 00:21:32,30.5830832 Clean data and process change are what? Turn AI from a demo into a result. 223 00:21:32,510.5830832 --> 00:21:38,300.5830832 People plus AI teams outperform either alone, but only if you actually redesign the work. 224 00:21:38,660.5830832 --> 00:21:50,480.5830832 And fifth, build trust by design guardrails, monitoring clear policies and AI literacy, especially teaching people when not to use AI are now the bottleneck. 225 00:21:52,752.4965082 --> 00:21:56,922.4965082 If you take nothing else from CES 2026, take this. 226 00:21:57,342.4965082 --> 00:21:59,262.4965082 AI is becoming infrastructure. 227 00:21:59,622.4965082 --> 00:22:03,882.4965082 That means it's going to be boring and expensive and absolutely worth getting. 228 00:22:03,882.4965082 --> 00:22:09,372.4965082 Right? And if you're feeling overwhelmed, good, that's a normal reaction to an infrastructure shift. 229 00:22:09,822.4965082 --> 00:22:12,852.4965082 The trick is to start with the boring wins and build from there. 230 00:22:13,542.4965082 --> 00:22:14,442.4965082 Thanks for listening. 231 00:22:14,562.4965082 --> 00:22:16,872.4965082 I'm Andy, and this was the AI breakdown.