AI Trends 2026: Quantum, Agentic AI & Smarter Automation

0:00 What will be the most important trends in AI in 2026?

0:03 Well, we take a stab at this every year with with some success, I would say.

0:08 And this time out, I have the knowledgeable assistance of my colleague,

0:12 Aaron Baughman, to help us out.

0:14 Well, yeah.

0:15 You know, after your prediction of infinite memory last year,

0:17 I thought maybe you could use just a little bit of help.

0:20 Yeah, that's that's fair.

0:21 Well, how about we each take four trends each?

0:24 That sounds good.

0:25 How about you first?

0:26 All right.

0:28 Okay.

0:28 So my number one trend of 2026 is multi-agent orchestration.

0:35 Now last year we said 2025 was the year of the agent.

0:40 AI agents that can reason and plan and take action

0:43 on a task and agents I think it's fair to say really delivered.

0:46 There are new numerous agentic platforms for tasks like coding

0:50 and basic computer use but no single agent really excels at everything.

0:55 So, what if you had a whole team of agents working together?

0:59 So, maybe we've got an agent here that kind

1:02 of acts as a planner agent that decomposes goals into steps.

1:07 Maybe we have some worker agents here that do

1:11 different steps like one specializes in writing code,

1:15 others call APIs and so forth.

1:17 And then perhaps we have a critic agent that evaluates outputs and flags issues.

1:24 And these agents collaborate under

1:28 a coordinating layer that is the orchestrator.

1:33 And multi-agent setups like this help

1:35 introduce cross-checking where one agent checks the other agents work and it can

1:41 break problems into more discrete verifiable steps.

1:44 Well, great.

1:45 So, how could I really follow that trend?

1:47 Well, I think I might just have one.

1:49 So, the second one is going to be the digital labor workforce.

1:53 So now these are digital workers that are

1:56 autonomous agents that can do a couple of items.

1:59 So the first one is they can parse a task by interpreting multimodal input.

2:04 So after preparation the worker then executes what's called a workflow.

2:08 Now this is where at the end of an action

2:12 plan you know it would follow a sequence of steps

2:15 but then it has to be integrated into some

2:18 sort of system that then in turn can take action.

2:22 And these could be downstream components.

2:24 Now these systems are then further

2:26 enhanced by what we call human-in-the-loop AI,

2:28 which then provides a couple of items.

2:31 The first one would be oversight.

2:33 The next one would be correction and then

2:35 we're looking at these strategic guidance or these rails

2:38 um to ensure that all of these agents

2:40 are doing what they're supposed to be doing.

2:42 Now this overall trend will create

2:44 a force multiplying effect to extend human capability.

2:48 Now trend number three is physical AI.

2:52 Now we all know that large language models they generate text like ABC.

3:00 And then there are other models as well.

3:02 So for example there are plenty

3:04 of diffusion image models and they generate pixels.

3:08 They generate images.

3:10 These are all operating in digital space.

3:13 Now, physical AI is about models that understand and interact

3:17 with the world that we live in, the the real 3D world.

3:22 And this is about models that can perceive their environment,

3:27 reason about physics, and that can take physical action like robotics.

3:33 So, previously getting a robot like

3:36 this to do something useful meant programming explicit rules.

3:40 So if you see an obstacle, you should turn left, for example.

3:44 And it was all done by humans.

3:48 It was up to yeah, smart guys like this to code these rules.

3:54 Now, physical AI kind of flips that around.

3:58 So you train models in simulation that simulate the real world

4:04 and it learns to understand how objects behave in the physical world,

4:08 how gravity works, how to grasp something without crushing it.

4:13 Now these models are sometimes called world foundation models.

4:18 They're generative models that can create and understand 3D environments.

4:23 They can predict what happens next in a physical scene.

4:27 And in 2026, many of these world models are

4:30 taking things like those humanoid robots that you found there,

4:34 Aaron, and they're taking them from research to commercial production.

4:38 Physical AI is scaling.

4:40 Well, Martin, you just took my trend,

4:43 but let's just go ahead and say number four is about social computing.

4:47 Now, this is a world where many agents

4:50 and humans operate within the shared AI fabric.

4:53 So say if I have an agent here and then a human here.

4:57 So they're going to be connected through this fabric

5:01 and here if I have information that flows between the two,

5:04 they begin to understand each other and then they

5:07 can gather what the intent is going to be.

5:10 And then once they have the intent and information, they have actions.

5:14 They can affect each other or maybe even the environment of which they're in.

5:18 But all of this flows seamlessly across this system.

5:22 It's this shared space that enables collaboration,

5:25 context exchange as well as event effective understanding.

5:28 Now the outcome is really an empathetic emergent network of these interactions.

5:32 It's what we call this collective

5:35 intelligence or this real world swarm computing.

5:38 So teams of agents, digital labor, humanoid robots,

5:42 and tech that can understand me with effective computing.

5:46 2026 could be uh quite the year and we're only halfway through the trends.

5:52 So trend number five that is verifiable AI.

5:58 Now the EU AI act is coming and by mid 2026 it becomes fully applicable.

6:06 And think of this a little bit like GDPR but for artificial intelligence.

6:11 Now, the core idea here is that AI systems, especially high-risk ones,

6:17 need to be auditable and they also need to be traceable.

6:20 Now, what does that mean?

6:22 Well, it means a few things.

6:24 It means documentation.

6:25 So, if you're building high-risk AI,

6:28 you need technical docs that demonstrate compliance to how

6:31 you tested the models and the risks that you identified.

6:35 It means transparency.

6:36 So, users need to know when they're interacting with the machine.

6:41 So things like synthetic text,

6:43 they need to be clearly labeled and it means data lineage.

6:47 You need to be able to summarize where your training

6:51 data came from and prove you respected copyright optouts.

6:54 And just like how GDPR has shaped global privacy, not just folks in the EU,

7:00 the EU AI act will probably set the template for AI governance worldwide.

7:04 Wow, that's great.

7:05 And you know, trend number six, right?

7:08 It really changes everything, but it also changes nothing at the same time.

7:13 And now this is where we put in quantum utility everywhere.

7:17 So 2026 is where we start to see this quantum

7:20 computing to reliably start solving real world problems better,

7:25 faster, or more efficiently than classical computing methods.

7:28 Now, at this point, we have this quantum utility scale.

7:32 is these systems that begin working alongside and together

7:35 with classical infrastructure to deliver

7:37 these practical value in everyday workflows.

7:40 Now, this is going to help with optimization

7:43 and then we'll also look at simulation and decision-making.

7:46 Now, all three of these tasks were

7:49 previously out of reach within the classical realm.

7:52 But this hybrid quantum classical error,

7:54 it will begin to transform quantum computing into this mainstream paradigm

7:58 as it's going to be woven into our everyday business operations.

8:03 Now my trend number seven is reasoning at the edge.

8:08 Now last year, we talked about very small models,

8:11 models with just a few billion parameters

8:13 that don't need huge data centers to run.

8:16 They work on your laptop or well maybe even your phone.

8:20 Well, in 2026, those small models are learning to think.

8:23 So, if we think about the best models that we have today, the frontier models,

8:29 well, pretty much all of them now use something called inference time compute.

8:36 They spend extra time thinking before giving you an answer,

8:40 working through problems step by step.

8:42 Now, the trade-off for that is they need more compute.

8:47 But here's what's changing.

8:49 Essentially, teams have figured out how they can

8:54 distill all of this reasoning information into smaller models.

8:59 So now these smaller models can perform thinking as well.

9:03 You're taking massive reasoning models that generate

9:06 tons of step-by-step solutions and we're

9:09 using that data to train the smaller models to reason the same way.

9:13 And that's resulting in reasoning models with only a few billion parameters.

9:18 They work offline.

9:19 Your data never leaves your device.

9:21 And there's no roundtrip latency to a data center.

9:24 So for anything that's real time or mission critical,

9:27 having a model that can actually reason through

9:30 a problem locally is a pretty big deal.

9:33 Yeah.

9:33 So that's all very true, Martin.

9:35 But now our last and final trend is number eight.

9:39 So this is what we're calling amorphous hybrid computing.

9:42 So this is a future where both AI model topologies and the cloud infrastructure,

9:48 they blend into what's called a fluid computing backbone.

9:51 So AI models, they're shifting beyond just this pure transformer design, right?

9:56 They're beginning to evolve into these other architectures that integrate

10:00 transformers and we call them these state space models.

10:04 And then in 2026, you're also going to see different emerging algorithms

10:09 that are combine both the state space

10:11 and transformers and other elements together, right?

10:14 And that's going to be really fun to watch, very artful.

10:18 And then at the same time, we have this cloud computing piece that's becoming

10:23 fully differentiated by combining many different chip types.

10:26 So we're going to have CPUs, GPUs, TPUs as well.

10:31 And finally, what we just talked about in trend six,

10:35 quantum, we're going to have QPUs.

10:38 I did also want to mention and note that you'll see

10:41 these neuromorphic chips that are coming out and those emulate the brain.

10:45 But all of these are going to be put together right

10:48 into this unified compute environment where parts

10:50 of each of these types of models,

10:53 they're going to be automatically mapped to the optimal compute substrate.

10:56 And this is really going to help

10:59 to deliver this maximum performance and efficiency.

11:01 And you know what?

11:02 Who knows?

11:03 But at this pace, probably not in 2026, but I think further out,

11:07 you might see DNA computing entering into the mix.

11:10 Well, those are some lofty goals.

11:12 And look, these are what we think are some of the biggest AI trends in 2026.

11:19 But what are we missing?

11:21 Which AI trend do you expect to be a big deal in 2026?

11:26 Yeah, let us know in the comments below.