0:01 Here are the six AI trends that will
0:04 matter most in 2026. This list is based
0:05 on daily research and reports from
0:07 institutions like McKenzie, OpenAI,
0:08 Stanford, [music]
0:10 and from analysts who are much more
0:12 knowledgeable than myself. In other
0:14 words, don't blame me if they get it
0:15 wrong. For each trend, I'll first start
0:17 with a big picture, then move on to the
0:19 actionable takeaways so that by the end,
0:21 you have a clear sense of where AI is
0:23 heading and what to [music] do about it.
0:24 Let's get started. Kicking things off
0:26 with trend number one. Models don't
0:28 matter much anymore. For the past few
0:29 years, every new model released sparked
0:32 debate about the best AI, and for good
0:33 reason. The difference in quality
0:36 between models was significant. In 2026,
0:38 though, that choice is going to matter a
0:40 lot less. Taking a look at the data,
0:42 this graph from artificial analysis
0:44 shows how the major AI models have
0:46 improved over time. Notice the
0:48 clustering in the top right corner. The
0:50 models are still getting smarter in
0:52 absolute terms, but the gap between them
0:55 keeps shrinking, meaning no single model
0:57 has a clear lead anymore. A Stanford
0:59 study confirms this from another angle
1:01 by comparing closed models like Gemini
1:04 and Chachi BT against openw weight
1:06 alternatives like Deep Seek and Llama.
1:08 The trend is pretty clear. Models that
1:09 are free to run are now approaching
1:12 frontier performance and performance is
1:15 only half the story. The cost matters as
1:17 well. Data from Epoch AI shows that
1:19 using powerful models has become
1:21 drastically cheaper and one of the
1:23 reasons is because hardware is getting
1:25 more efficient. For perspective,
1:29 Nvidia's latest chips uses 105,000
1:32 times less energy per token than they
1:34 did 10 years ago. So, what does this
1:36 mean for us? In plain English, when
1:38 things get cheaper and more similar,
1:40 they become more like commodities. You
1:42 don't ask who provides the best
1:44 electricity, right? You ask what can I
1:46 use the electricity for? And because of
1:48 this, the competition is shifting from
1:50 the AI model itself to the way we
1:53 actually use it, aka the app layer. Just
1:55 think about cars. Once the engine
1:57 becomes standardized, the focus shifts
1:59 to the features and the design. This
2:01 creates an interesting dynamic for each
2:03 of the frontier AI labs. For example,
2:05 OpenAI has a mind share advantage
2:07 because ChachiBT is synonymous with AI
2:10 and has the largest market share. Google
2:11 has a distribution advantage because
2:14 Gemini is already embedded across its
2:15 existing products like search, Gmail,
2:17 and Android. Anthropic has a
2:19 specialization advantage given its loyal
2:21 customer base in developers and
2:23 enterprise customers. Notice what's
2:24 missing from that list. None of them are
2:27 winning because they have the best AI.
2:29 The competition has moved beyond raw
2:31 power to reach, integration, and trust.
2:33 The practical takeaway here is to stop
2:35 obsessing over technical scores and
2:37 instead focus on how they fit into your
2:39 actual work. For example, if you live in
2:41 Google Workspace, Gemini's deep
2:43 integration with all of Google's apps
2:45 gives it an edge that has nothing to do
2:48 with raw intelligence. By the way, I'll
2:49 link all the sources I mentioned today
2:51 down below so you can check them out for
2:53 yourself. Trend number two, 2026 is the
2:56 year of AI workflows, not AI agents. If
2:58 you spend any time on Twitter or
2:59 LinkedIn, you've probably noticed the
3:01 industry jump from chat bots straight to
3:03 autonomous agents and completely skip
3:05 the middle step where the actual value
3:08 is being unlocked, AI workflows. And the
3:10 numbers prove this. According to
3:12 McKenzie, no more than 10% of
3:14 organizations in any given business
3:17 function report scaling true agents.
3:19 Meanwhile, we see from OpenAI's
3:22 enterprise report that 20% of enterprise
3:24 AI use is already happening through
3:27 workflow specific tools like custom GBTs
3:28 and projects. [music]
3:30 This gap tells you the market has voted
3:32 for workflows, not autonomy. And we're
3:34 seeing this play out across industries.
3:36 A pharma company redesigned their
3:38 clinical study workflow by using AI to
3:41 analyze raw clinical data while humans
3:44 focus on validation leading to 60% less
3:48 prep time and 50% fewer errors. A
3:49 utility company redesigned their call
3:51 center workflow where AI handles
3:53 authentication and routine inquiries
3:55 cutting cost per call by 50% while
3:58 increasing satisfaction scores by 6%. A
4:00 bank redesigned their code migration
4:02 workflow where AI scans legacy code and
4:04 generates updated versions for
4:06 developers to verify, cutting the
4:08 required human hours by 50%. Andre
4:10 Kaparthi sums it up perfectly, calling
4:12 everything an agent creates unrealistic
4:14 expectations and confusion. Fully
4:16 autonomous AI still faces massive
4:18 hurdles like data security. So, we're
4:21 looking at the decade of agents, not the year.
4:22 year.
4:23 >> I was triggered by that because I feel
4:25 like there's some overpredictions going
4:28 on in the industry. And uh in my mind
4:29 this is really a lot more accurately
4:31 described as the decade of agents.
4:32 >> Meanwhile, by integrating something like
4:35 custom GBTS into an existing workflow,
4:38 we've essentially created an agent light
4:41 which is much more reliable at producing
4:43 consistent results. To really ram this
4:45 point home, McKenzie predicts that
4:47 redesigning workflows will unlock nearly
4:50 $3 trillion in economic value by 2030.
4:52 And more importantly, these
4:53 organizations will have the muscle
4:56 memory to adopt true AI agents faster
4:58 when they finally arrive. So here's your
5:00 practical takeaway. Your goal for 2026
5:02 is to turn your successful prompts into
5:04 repeatable workflows. And this is
5:05 something I've talked about in other
5:07 videos. Pick one recurring deliverable
5:09 you produce, like a weekly report. Break
5:12 it into steps and let AI handle the
5:14 predictable parts. Keep yourself in the
5:16 loop for the final judgment calls. That
5:18 structure is what creates true
5:19 reliability. Side note, I'm actually
5:21 developing an entire course around
5:24 evergreen AI skills to give you a future
5:25 proof framework that never becomes
5:26 obsolete. If you're interested in
5:28 learning a practical and timeless AI
5:30 system, click the link below to join the
5:32 wait list. Trend number three, the end
5:34 of the technical divide. When I was at
5:36 Google, non-technical teams like sales
5:38 and marketing had to rely on specialist
5:39 teams to help them build stuff like
5:41 dashboards. And I'm not someone who
5:43 holds grudges, but a lot of my requests
5:45 were depprioritized because they were
5:48 too low impact and my clients weren't
5:50 key accounts, but no, I'm over it.
5:52 Anyways, in 2026, that's going to happen
5:54 a lot less. The numbers backing this are
5:56 honestly kind of insane. According to
5:58 Open Eyes latest report, 75% of
6:01 enterprise users reported using AI to
6:03 complete tasks they literally could not
6:05 do before. Not just doing old tasks
6:07 faster, they're doing entirely new
6:09 things. For example, coding related
6:11 messages from non-technical employees
6:14 grew 36% in just 6 months. These are
6:16 salespeople, marketers, and operations
6:18 managers writing scripts, automating
6:20 spreadsheets, and building internal
6:23 tools. A study from MIT confirms this.
6:25 AI acts as an equalizer,
6:27 disproportionately helping workers with
6:29 lower technical skills close the
6:31 performance gap with [music] experts.
6:32 Here's what all this means for your
6:34 career. If your value is purely
6:36 technical, aka you're the dashboard
6:38 person, then your competitive advantage
6:39 is shrinking because the marketing
6:41 manager who used to wait in your queue
6:43 can now do it themselves. [music] But if
6:45 you are that marketing manager or the
6:47 salesperson who deeply understands their
6:48 clients, then this is the biggest
6:50 opportunity of your career because the
6:52 technical barrier that stood between
6:54 your expertise and your execution is now
6:56 gone. Here's your practical takeaway.
6:59 Attempt one impossible task this month.
7:01 Identify a technical project you usually
7:02 outsource like building a dashboard,
7:04 cleaning a messy data set, or automating
7:07 a report and try doing it yourself using
7:09 Gemini Cloud or Cashibbt. You'll be
7:11 surprised by what you can now pull off
7:12 alone. Moving on to trend number four,
7:15 from prompting to context. One of my
7:16 most popular videos is this one teaching
7:18 you how to prompt because as we all
7:20 know, if we don't phrase our request
7:22 well, we get a bad result from AI.
7:23 Unfortunately for me, that video is
7:25 going to matter a lot less in 2026
7:27 because new models have gotten so much
7:29 better at understanding vague
7:31 instructions. However, they still have
7:33 one massive weakness I call the fact
7:34 gap. While models know almost everything
7:36 on the public internet from Shakespeare
7:39 to Python code, they know nothing about
7:41 your Q3 goals, your brand guidelines, or
7:43 that email your boss sent yesterday.
7:45 It's like having a brilliant employee
7:47 who technically knows how to complete
7:49 tasks, but isn't allowed to look at any
7:51 company files. they're still going to
7:53 fail, right? Because they lack context.
7:54 At least that's what I told my boss
7:56 during my first internship. It's the
7:58 exact same thing with AI. The focus has
8:01 shifted from how we ask the wording to
8:03 what we give it, the context. And this
8:05 explains the platform wars we're seeing
8:07 right now. Google, Microsoft, and others
8:09 are racing to embed AI into their
8:11 productivity suites because whoever
8:13 holds your context, your emails, your
8:15 docs, your calendar, holds your
8:17 attention. This is also how they'll trap
8:19 you with platform lockin. The more
8:21 context you build up in one ecosystem,
8:23 the smarter the AI is for you and the
8:25 harder it becomes to switch. There are
8:26 two practical takeaways here and the
8:28 non-productivity people are going to
8:30 hate this. First, file management is no
8:32 longer optional. To get value from AI,
8:34 you need some sort of system to keep
8:36 your files organized and clearly named.
8:38 If your work is scattered in random,
8:40 unnamed folders, you can't point the AI
8:42 to it. Second, audit where your
8:44 information lives. If it's spread across
8:46 three or four different platforms, you
8:47 need to consolidate. If your resume
8:49 lives in Google Drive, but the job
8:50 description and interview notes are
8:53 stored in Notion, neither Gemini nor
8:55 Notion AI can help with interview prep,
8:57 you end up doing the synthesis manually,
8:59 which leads to more friction and defeats
9:01 the whole purpose. So, as a rule of
9:03 thumb, prompting still matters, but it's
9:05 more important to ask yourself, does the
9:08 AI have the files it needs to know what
9:10 I'm talking about? Trend number five,
9:12 advertising is coming to chat bots, and
9:14 it's not all bad. First of all, please
9:15 don't shoot the messenger on this one.
9:17 Hear me out. At this point, it's
9:18 basically been confirmed that ads are
9:20 coming to CHACHBT in 2026. So, instead
9:22 of debating if it will happen, let's
9:24 talk about the implications. Imagine a
9:27 world where advertising never comes to
9:29 chatbots. In that reality, the best AI
9:31 models stay locked behind expensive
9:34 subscriptions, creating a wealth gap,
9:36 where only those who can pay have access
9:38 to the best tools, while everyone else
9:41 is left with an inferior version. Over
9:43 time, this creates a compounding
9:45 advantage. The wealthy use powerful AI
9:47 to get wealthier while everyone else
9:49 falls further behind. Kind of reminds me
9:52 of something I just can't put my finger
9:55 on. It think of it like YouTube. Imagine
9:56 if you couldn't watch videos from the
9:58 top creators unless you pay for YouTube
10:00 Premium. That is where AI is headed
10:02 without an ad supported tier. Now that
10:04 we know ads are inevitable and that I'm
10:06 not to blame for this, uh the thing to
10:08 watch is what format those ads will take
10:10 because it's going to look very
10:12 different from the search ads we're
10:13 currently used to. For example, industry
10:16 expert Eric Sufer predicts chatbot ads
10:18 will not be tied to our specific
10:20 questions because if an AI recommended a
10:22 product directly in its answer, we
10:24 wouldn't trust it. Instead, the ad will
10:25 probably look like standard display
10:27 banners that stay separate from your
10:29 actual conversation. Sort of like the
10:31 banner ads we see on websites today. So,
10:32 here's the bottom line. I don't like
10:34 ads. You don't like ads. Nobody likes
10:36 ads. But it's the ad revenue that makes
10:38 it possible for companies to offer their
10:40 best models to students in developing
10:43 countries, nonprofits, and casual users
10:45 who can't afford another monthly bill.
10:47 Trend number six, from chatbots to
10:49 robots. Everything we've covered so far
10:51 has focused on AI as software. But in
10:54 2026, that software is going to appear
10:56 even more in the physical world as
10:58 physical agents who can move on their
11:00 own. The numbers show this is already
11:02 happening. Exhibit A, Whimo. Their
11:04 autonomous taxi service has now logged
11:07 over 100 million fully autonomous miles
11:10 and are involved in 96% fewer crashes
11:13 than human drivers. Exhibit B, Amazon.
11:15 Their AI enabled warehouse robots have
11:17 cut the time from order to shipping by
11:21 78%. Exhibit C, China. As early as 2023,
11:23 China had deployed more industrial
11:26 robots than the US and the rest of the
11:28 world combined. Now, there is one caveat
11:31 to all this. Humanoid robots are still
11:39 MIT robotics professor Rodney Brooks
11:42 estimates that we are at least 15 years
11:44 away from seeing functional humanoid
11:46 robots in our daily lives. The real
11:48 shift is what analyst Mary Miker calls
11:51 AI turning capital assets into software
11:53 endpoints. And here's what that means in
11:55 plain English. A car, a tractor, or
11:57 warehouse robot used to be a
11:59 depreciating asset, which means it loses
12:02 value as time goes on. Right now, these
12:03 machines are becoming platforms that
12:05 improve over time through software
12:08 updates, exactly like our phones. A
12:11 Whimo car today is actually safer and
12:13 smarter than it was 2 years ago, even if
12:16 the physical vehicle hasn't changed. So,
12:18 what does all this mean for us? In a
12:20 nutshell, while the headlines are
12:22 focusing on white collar disruption for
12:24 now, this trend suggests that blue
12:26 collar work will also be disrupted, but
12:29 over a much longer time horizon. On a
12:30 more positive note, I want to leave you
12:32 with something Ethan Mollik said. He's a
12:33 professor at Wharton, and this is
12:34 something I really believe in. His
12:36 research on what he calls the jagged
12:39 frontier of AI shows that right now we
12:42 are in a unique window where expertise
12:44 is being reset thanks to AI. And
12:46 precisely because things are messy and
12:49 undefined right now, there are no
12:51 experts who know everything already. You
12:53 just need to be willing to learn faster
12:55 than the person next to you. That is how
12:57 you win in 2026.
12:59 Stop worrying about developing a perfect
13:02 plan to learn AI and instead just get
13:03 started. I'd love to hear your thoughts
13:05 on these trends, so drop them down
13:06 below. Check out this practical guide on
13:08 Google Gemini next. See you on the next
13:10 video. In the meantime, have a great one.
