Latest / In The Loop / Nine AI Trends For 2026

Transcript

Nine AI Trends For 2026 · In The Loop

• 0:00
  At the start of every single year, I think it's.
• 0:03
  Important. To provide what we believe is
• 0:06
  going to be the trends that. Shape the next 12 months of AI.
• 0:10
  2025 was a year of huge improvement in what?
• 0:14
  AI can actually deliver for people.
• 0:16
  However, I think. Expectations are still.
• 0:20
  Outstripping reality, 2026 is going to be another year of huge
• 0:24
  progression, and in today's episode, we're going to talk
• 0:26
  about the top. Trends that are going to shape
• 0:29
  the world of AI this year, there's no doop with Jack
• 0:33
  Harton. Hope you enjoy the show.
• 0:47
  So we're actually 21 days into January.
• 0:50
  I was going to take a couple of. Weeks off over Christmas to have
• 0:52
  a break. Then was moving house.
• 0:56
  Lost a bunch. Of things.
• 0:57
  And even today, we're actually recording.
• 0:59
  I've still not got the, let's say the studio to, so to speak,
• 1:02
  set up. The actual SD cards are
• 1:04
  somewhere still living in boxes from the Move, so I'm going to
• 1:08
  apologise me wearing a silly headset because actually it was
• 1:10
  probably the best microphone I could possibly get I think.
• 1:14
  Hopefully. We're going to use an AI
• 1:15
  software to make it sound as good as we possibly can.
• 1:29
  So now anyway. Moving on to.
• 1:30
  Trend number. One which I'm going to describe
• 1:32
  in. A couple of words which is.
• 1:35
  Differentiation for companies. So ultimately, right now,
• 1:39
  differentiation 12 months. Ago in the AI space used.
• 1:42
  To look a. Lot different.
• 1:43
  To what it. Looks like today because
• 1:46
  fundamentally the way users interact.
• 1:48
  With AI is just converging on the exact.
• 1:50
  Same pattern everywhere. Really.
• 1:51
  You know every company building an AI product is.
• 1:55
  Essentially a chat bot. You know, you type something,
• 1:58
  the AI responds. There's maybe interactive
• 2:00
  elements that pop up and appear. You know you've got Amazon's ad.
• 2:03
  Management software's the exact same as Google.
• 2:06
  'S and you know every single. Interface just starts to look
• 2:09
  quite similar with unique. Parts of branding and.
• 2:12
  Colours and vibe as well. But mostly it's a chat interface
• 2:16
  and you know it makes. Sense there's over. 1 1/2
• 2:18
  billion people. Using AI every single day right
• 2:20
  now. And you know that that.
• 2:23
  Expectation is transferred to every product.
• 2:25
  So everyone, naturally. Is becoming a chat.
• 2:27
  Bot and it's. Therefore, quite difficult as
• 2:30
  organizations, as people to go. What is it that makes this
• 2:33
  company unique Then? You know, if everyone looks the
• 2:35
  same and it's an agent doing tasks for people, what makes
• 2:38
  them, you know, unique? So what's happening is really
• 2:41
  interesting. Questions being posed, which is
• 2:43
  OK, therefore. What isn't isn't
• 2:45
  differentiation. That's going to be a big
• 2:46
  question of this year. My belief is.
• 2:49
  It's not the interface anymore, it's 12 months ago strategically
• 2:53
  companies building. New chat interfaces and
• 2:55
  experiences that was. Strategic, clever work.
• 2:58
  Now it's just. Not again, because building
• 3:01
  better chat interfaces is really important, but doesn't
• 3:03
  differentiate. Anybody you know, everyone has
• 3:06
  to be solving that exact same problem the exact same way.
• 3:09
  And you? Everyone has it, every user
• 3:11
  expects it, so it becomes table stakes, so therefore it raises.
• 3:15
  An interesting question if the. Chat interface or the interface
• 3:17
  layer that used to be for companies, a place that offered
• 3:20
  massive differentiation. You know, big companies like
• 3:22
  let's say Salesforce or HubSpot have ecosystems and training
• 3:26
  programs and certificates for becoming a master at my
• 3:29
  software. They used to literally monetize
• 3:32
  friction of having lots of things and features on a page
• 3:36
  because a salesperson could easily point it in and say,
• 3:38
  look, we do everything and it doesn't.
• 3:40
  Matter that you don't know how because we've got big training
• 3:42
  programs to make. You an expert?
• 3:44
  So that used to be differentiation because you used
• 3:46
  to. Look.
• 3:46
  Complex, whereas now everything is a.
• 3:49
  Chat interface and an agent orchestrates and just takes all
• 3:52
  those actions. So therefore it's a really
• 3:53
  interesting one because. The question becomes, where does
• 3:56
  competitive advantage therefore come from and.
• 3:58
  This is where I'm going to go to trend #2 which?
• 4:00
  Is where companies are going to start investing effort.
• 4:03
  So trend 2. Broadly is is where?
• 4:05
  Companies are redirecting. Their efforts towards what let's
• 4:09
  describe as their unique intelligence.
• 4:11
  So what? Do I?
• 4:12
  Mean by intelligence? So the intelligence.
• 4:14
  Is about unique ways of solving customer problems in your
• 4:20
  specific domain. So for example, if you're
• 4:22
  building, let's say. For example, a company that is
• 4:25
  selling sales software. Which is maybe they're going to
• 4:28
  expose. Unique intelligence, which is
• 4:30
  the ability to, say, predict which deals might close based on
• 4:34
  five years of deal history. And maybe showing patterns of
• 4:38
  historic. Alternatives and competitors and
• 4:42
  behaviours from salespeople that have led to more deals being won
• 4:46
  or less deals and that being exposed for an agent and a user.
• 4:50
  So that's intelligence. So if it's a logistics platform,
• 4:53
  so it helps people get packages from A-Z.
• 4:56
  Maybe they can. Analyse millions of shipments.
• 4:59
  Across every region in the world.
• 5:01
  And then expose that to the customer through AI or if your
• 5:04
  customer support platform for example.
• 5:07
  It's understanding exactly. Which?
• 5:08
  Issues really need human. Intervention based on historic
• 5:11
  data. So that's what we mean by
• 5:14
  intelligence. And that's where.
• 5:15
  Companies, we believe. Over this next period is going
• 5:18
  to be. Trickling down on.
• 5:19
  Because if they expose more and more and more of that.
• 5:22
  That is extremely. Unique and something no one else
• 5:24
  has. Rather than spending all that
• 5:26
  time on the chat interface, it's just about exposing that.
• 5:29
  Value to. Customers and users.
• 5:31
  So yeah, I think that's going to be a massive trend of companies
• 5:34
  trying to. Expose that very unique
• 5:35
  intelligence, rather. Than think through.
• 5:37
  Long user journeys and experiences anymore because
• 5:39
  ultimately. The chat interface all looks.
• 5:42
  The same. It's the outcomes an agent.
• 5:45
  Can deliver for a person and how clever those outcomes are.
• 5:48
  That is what delivers value in the future anyway.
• 5:50
  We're going to go to trend #3. Trend #3 is that.
• 5:53
  AI conversations are about to. Become immensely visual.
• 5:56
  So yeah, AI conversations are moving from like text.
• 5:58
  Based to. Really visual, interactive.
• 6:01
  Experiences. So you've seen this with Chatchi
• 6:02
  BT? So Open AI and Chatchi BT
• 6:05
  released integrations into Spotify, Expedia, Instacart, and
• 6:08
  essentially the Spotify playlist renders in a conversation.
• 6:12
  Or you can do learning. Through the big learning
• 6:15
  platforms straight into track GBT and their platform appears
• 6:18
  inside the chat interface. So that what's becoming a trend
• 6:22
  really is that the chat interface.
• 6:24
  Wants to render beautiful user interface components from
• 6:27
  different platforms because fundamentally again.
• 6:29
  The more outcomes and agent. Can deliver, the more valuable
• 6:32
  it is, and so for all companies they want more.
• 6:35
  UI. Components deliver more outcomes
• 6:38
  in their chat. Because if you think about,
• 6:40
  let's say. I want to select a playlist that
• 6:43
  might have taken a conversation that lasted 10 minutes suddenly.
• 6:46
  The user interface. Visual component appears in the
• 6:48
  chart and you just select your playlist and press go.
• 6:52
  So you want to book a flight. Instead of reading a description
• 6:54
  of flight information, you just have a comparison of flights in
• 7:00
  a visual format that has ticks and crosses.
• 7:02
  For example, just as you would get on a normal website that's
• 7:04
  straight from ChatGPT or straight from a conversation.
• 7:07
  So you can bet. £1,000,000 that. Claude is about to do the same.
• 7:11
  Microsoft, Apple, when they actually get their act together.
• 7:14
  A slightly technical diversion here is that the model context
• 7:17
  protocols, so MCP protocol. So MCP is actually expanding
• 7:23
  what it enables you to do. So actually you can provide UI
• 7:26
  components inside of that very soon, which means that as one
• 7:30
  vendor to another or an organization, you can then let's
• 7:34
  say you sell your maps software, you can not just put your maps
• 7:38
  information into an MCP, you can put your maps user interfaces
• 7:42
  into an MCP and then other platforms might go to surface
• 7:45
  that up through a conversation. So yeah, that's.
• 7:47
  The three trends on. Let's describe it as kind of the
• 7:50
  interface part, so where the AI interface is changing.
• 7:55
  I want to make a change of direction and move on to I guess
• 7:59
  trends on how. Agents are built.
• 8:12
  So trend #4. Of this list I think.
• 8:14
  Agent reasoning is. Going to start to really
• 8:16
  squeeze. And compete.
• 8:18
  Rigid workflows or even clever workflows from the likes of
• 8:21
  Zapier NAN. Lindy.
• 8:24
  So the very big popular platforms for AI automation.
• 8:27
  And I think really there's a. Big debate going on online all
• 8:29
  the time. Which is you see a lot of
• 8:32
  people. Sharing their great NAN workflow
• 8:34
  that saved ME3. 1000 hours of time.
• 8:36
  People see that, they play with it and it works, It makes
• 8:38
  complete sense, and for many tasks it makes a lot of sense to
• 8:41
  use that. But as a result, people assume
• 8:43
  that's the way agents should be built and LED, and that's where
• 8:46
  the fundamental debate comes in and you've got the likes of Lang
• 8:49
  Chen. CEO who really complain that a.
• 8:52
  Open AI release yet another workflow builder.
• 8:56
  You've got anthropic complaining.
• 8:58
  About the exact same. Because they don't.
• 8:59
  Believe workflows are the future and so.
• 9:02
  A lot of. Companies default to workflows
• 9:03
  because they feel like they're able to.
• 9:06
  Be in control, so I. Know this agent is going to do
• 9:08
  this, then it's going to do that.
• 9:11
  But often when? It comes to real world
• 9:14
  complexity for, let's say, deploying workflows.
• 9:18
  They're going to break because a human person wants to go in 100
• 9:21
  different directions. And that's where agents who can
• 9:24
  reason and create a plan like you.
• 9:26
  Might see on cloud. So it creates a specific plan,
• 9:28
  it asks itself questions, it ticks itself off on a plan and
• 9:31
  then uses information to shape that plan.
• 9:34
  That's what actually a workflow is doing.
• 9:36
  But you're trying to build a workflow and try and plan for
• 9:39
  hundreds of scenarios, whereas an agent can do that and change
• 9:42
  it on the fly. So yeah, Zapier.
• 9:44
  NAN and tools like that. I definitely feel attractive
• 9:47
  because you know, you can drag and drop things, you can select
• 9:50
  arrows, you can define logic for an agent must do this and then
• 9:54
  that. You can literally see what's
• 9:56
  going to happen. See, as I said, but life.
• 9:59
  Tasks. And humans just don't like
• 10:02
  following that sequence of steps, so.
• 10:04
  If I wanted. To do a workflow, I might go and
• 10:06
  set a step for saying. Check for duplicates and if
• 10:09
  found. An issue?
• 10:11
  Refund them, then check cancellations.
• 10:13
  And if you find an issue, retry. Check the payments, but if
• 10:16
  there's an invalid prompt for update.
• 10:18
  But the duplicate. Charge was.
• 10:19
  Actually an upgrade the cancellation.
• 10:21
  Would then fail because there's a pending charge.
• 10:23
  So everything becomes. Interconnected.
• 10:24
  The workflow can't handle that level of interconnectedness and
• 10:27
  chaos. It becomes very difficult.
• 10:29
  Whereas agent reasoning can assess.
• 10:31
  A full situation. Have a conversation with the
• 10:33
  user. Identify the three issues.
• 10:35
  And know that they're. Related because it is able to
• 10:37
  analyse more information. And create a plan and again it
• 10:40
  can. Change its plan and then check
• 10:42
  in with the users Say is this good or bad?
• 10:44
  See, I think a massive trend will be moving towards agent.
• 10:46
  Reasoning because people who love cloud code and how powerful
• 10:49
  it is and how effective it is and it's something we're mindset
• 10:52
  tripling down on. Because I ultimately.
• 10:55
  We see the value in that. It's where we're just releasing
• 10:59
  new technology where I can look over millions of rows of data
• 11:02
  and create a plan for how to. Click into very.
• 11:05
  Complex questions and. Find the right data to help it.
• 11:08
  Answer those questions. And give real value to users.
• 11:12
  So trend 5. Is.
• 11:14
  Again, within the theme of how agents.
• 11:16
  Are built. And specifically, this theme is
• 11:18
  how I think Building Agents is going to move to.
• 11:21
  More conversational interfaces. And I think that's what's.
• 11:23
  Going to get widespread adoption.
• 11:25
  So I think it's going to move from quite technical
• 11:27
  configuration or even. Workflows or even.
• 11:30
  Code to configuration. Via conversations, you know
• 11:35
  describe what the agent should do and the system should just
• 11:37
  generate the configuration and this might expand who builds
• 11:40
  agents. From maybe 2%.
• 11:42
  Of people and knowledge workers to, you know, 50 percent, 60%.
• 11:46
  Because right now, building an agent.
• 11:48
  Requires quite a few technical skills and thinking a lot of the
• 11:50
  time, even if you try and simplify it.
• 11:53
  You know, Zapier and other similar tools.
• 11:55
  Make you think like a programmer.
• 11:58
  You have to understand variables and data types and condition
• 12:01
  logic and error handling and code first.
• 12:04
  Approach. Requires actual programming.
• 12:07
  And the issue there is that. An agent's very rarely one and
• 12:10
  done. You're then going to.
• 12:11
  Have to maintain it and improve. It and change.
• 12:13
  It So yeah, what I think is going to start emerging.
• 12:15
  For agent building is very conversational studio
• 12:17
  interfaces. It's something we're actually
• 12:19
  about to start launching as AV two of our agent.
• 12:22
  Building Google AI. Studio is now moving towards
• 12:25
  that where you can. Configure these experiences.
• 12:28
  Through pretty. Minimal.
• 12:31
  Interface, really, so soon you'll be able to say I just
• 12:33
  want an agent that monitors support tickets, figures out you
• 12:36
  know, which ones are urgent and just reach them to the right
• 12:38
  person. And if a ticket sits there for
• 12:40
  more than 24 hours, escalate it. That's how we think and in terms
• 12:43
  of instructions and the system should just know how to do that
• 12:45
  and build that trend 6, I think. Agent testing is about to become
• 12:49
  a major. Area of focus for all companies
• 12:53
  because the problem. Often isn't AI.
• 12:54
  Capability and This is why people gravitate, we think to
• 12:57
  workflows the. Problem is confidence.
• 12:59
  Confidence that agents are going to make good decisions
• 13:01
  repeatedly in scenarios. Actually kind of matter.
• 13:04
  Large scale AI LED training and testing is going to be what
• 13:07
  builds that confidence because traditional say software testing
• 13:11
  of a feature doesn't work Agent. Testing is quite different
• 13:13
  because. Behavior isn't, let's say,
• 13:15
  deterministic. Give an agent a support.
• 13:17
  Query. It might handle it in five ways
• 13:20
  over 5 separate occasions, even though it should be consistent.
• 13:23
  Multiple. Approaches could all be correct,
• 13:26
  but slightly different, which makes us all as humans feel
• 13:28
  quite nervous. It's like a human.
• 13:30
  And so the answer really for. Us is I think we're at mindset
• 13:33
  and what we're going to expect us to see everywhere in the
• 13:36
  world very soon, which is that often ahead of things by, you
• 13:39
  know, three to seven months is scenario based testing at
• 13:42
  massive scale using AI to evaluate everything.
• 13:45
  So you. Describe the scenarios that
• 13:47
  represent real scenarios or situations.
• 13:50
  Define success criteria based on judgement and use AI to.
• 13:53
  Test itself and. Test AI.
• 13:55
  And I think this is all. Going to connect to ROI as well
• 13:57
  because I think people are going to move from say vibes of AI
• 14:01
  deployments to real ROI and impact of performance.
• 14:15
  So that's how agents are. Billed as a theme set.
• 14:18
  I think let's move on now to another theme which.
• 14:20
  Is how work. Changes.
• 14:21
  As a result of all of this and as part of this.
• 14:23
  Let's go into. Trend 7, which is I think.
• 14:26
  Knowledge work is. Going to start.
• 14:27
  Shifting from doing. To now managing because all of
• 14:31
  these things that we're going to start to see is going to.
• 14:32
  Simplify agent creation. But make it more powerful
• 14:35
  because workflows and complex interfaces is not what's going
• 14:38
  to deliver wide scale, let's say, adoption amongst the
• 14:43
  workforce, especially in knowledge work.
• 14:46
  So I think all of these things are going to start culminating
• 14:49
  in knowledge workers moving from say 1 or 2% or 8% of people
• 14:52
  working really fast to let's say 50, sixty, 70%.
• 14:56
  You're starting to see this now with Anthropic launching Cloud
• 14:59
  Co work. Which is essentially.
• 15:01
  Trying to give the experience of Cloud Code to.
• 15:04
  Non-technical people because the number of non.
• 15:07
  Technical people using cloud code again, trying to build and
• 15:10
  provide an agent LED experience with a series of tools like I
• 15:12
  was describing. No workflows and making it
• 15:15
  really simple to manage for them.
• 15:17
  So I think knowledge work is going to shift from.
• 15:19
  Doing tasks to managing agents and therefore testing frameworks
• 15:23
  and simplicity is going to be really important within this
• 15:26
  theme, I think. Trend A is.
• 15:27
  Going to be. The year.
• 15:29
  Of agent manager. As a result.
• 15:30
  And the reason I say agent management, not full autonomy is
• 15:34
  because I don't think full autonomy is close.
• 15:36
  I think we're quite a while away from that.
• 15:37
  So I think that's why. Agents are so.
• 15:39
  Powerful versus, say, a workflow.
• 15:41
  Because an agent will. Do a series of tasks and then
• 15:43
  check in with you, ask the user to make that judgement call,
• 15:48
  make strategic decisions. But the agent will start to
• 15:50
  increasingly handle more and more of the execution because
• 15:53
  it's simpler just to provide agents tools and agents to
• 15:56
  reasoning. So yeah, I think This is why
• 15:58
  it's going to be so. Powerful The a lot.
• 15:59
  Of these trends are just going to culminate in some of these
• 16:02
  things because ultimately. Human judgements going to be.
• 16:04
  Still really important, but agents begin to do powerful.
• 16:07
  Things with powerful reasoning. Is going to be widespread across
• 16:10
  many different products, not just Chatter, BT and cloud.
• 16:14
  It's something that we're providing all of our customers.
• 16:15
  For example, the ability to have reasoning agents and the ability
• 16:19
  to run large scale testing on your agents and AI experiences.
• 16:23
  Another big theme I think we're going to start to see as Tread 9
• 16:26
  is forward deployed vibe coding. So.
• 16:29
  We we shouldn't actually. Use the word vibe coding because
• 16:32
  it almost, let's say, lowers the perceived value of using AI as a
• 16:36
  coding tool. But anyway, there's a there's a
• 16:38
  new role that's emerging more and more.
• 16:39
  Which is people in departments who come.
• 16:41
  Back who combine domain expertise with AI building
• 16:45
  capabilities essentially. So not necessarily professional
• 16:48
  engineers, but domain experts who can build solutions to
• 16:51
  problems so these. People know really how to use
• 16:54
  AI. Tools and the domain experts and
• 16:56
  they forward deploy them to solve big problems and not get
• 17:00
  it to let's say full production or ready for everyone to work on
• 17:04
  and or to deploy to customers or launch to customers.
• 17:07
  What we are seeing is they may have to solve big knotty
• 17:09
  problems because maybe they're expert coders and they have
• 17:11
  domain expertise or they're just good at AI tools and have
• 17:14
  fantastic AI and domain expertise.
• 17:17
  And the reason domain expertise is important is ultimately they
• 17:20
  have a lot a lot of context to make good judgement calls and
• 17:24
  solve big problems for companies.
• 17:26
  And that could be. Ideas for new customer products?
• 17:30
  It could be solving big problems for teams and essentially
• 17:34
  building tools and systems using AI and then eventually handing
• 17:38
  it over to the rest of the team. So these are the trends that we
• 17:51
  think will. Define AI in 2026.
• 17:54
  The interface is experiencing the biggest revolution since
• 17:59
  command line interfaces. Companies will be investing more
• 18:02
  and. More in their own data to
• 18:03
  provide as unique intelligence to their users and customers and
• 18:07
  ultimately. You'll see more.
• 18:08
  Deployments than ever throughout 2026.
• 18:10
  So that's it for this week in this episode.
• 18:12
  I hope you found it useful. Next week we'll have a proper.
• 18:14
  Microphone, camera and setup. But yeah, we'll see you next
• 18:17
  week.