Kind: captions
Language: en
I've been testing chat GPT's new agent
mode for a week and the results were
interesting. While everyone's focused on
what these features can do, I found some
important details about how they
actually perform that could save you
time and money. There are some things
about usage limits, performance speeds,
and reliability that you really need to
know before diving in. And honestly, the
reality is more nuanced than most
reviews are telling you. Welcome back to
Bit Biased, where we test AI tools with
our own money so you don't have to. In
this video, I'm showing you exactly what
works, what doesn't, and the hidden
costs nobody mentions. Plus, I'll walk
you through the setup process and share
the specific prompts that actually get
results. What just changed everything.
All right, so let's talk about what
OpenAI actually released here because
there's been a lot of confusion online
about what's new and what isn't. The big
star of the show is agent mode. And
guys, this is wild. Instead of chat GPT
just sitting there waiting for you to
ask questions, it can now actually go
out into the world and get things done.
We're talking about AI that opens
websites, clicks buttons, fills out
forms, creates actual documents you can
download, and even manages your
calendar. But here's where it gets
really interesting. They also rolled out
the 03 Pro model, which is basically
chat GPT on steroids for complex problem
solving. Then there's the new GPT4.1
with a million token context window.
That's like feeding it an entire novel
and having it remember every detail. And
honestly, the more I've been playing
with these features, the more I realize
we're looking at a completely different
category of AI tool here. This isn't
just a smarter chatbot. It's like having
a really capable intern who never gets
tired and works at lightning speed. Let
me show you exactly what I mean. Agent
mode, your digital employee. So, agent
mode is probably the most mind-blowing
feature here, and I've been putting it
through its paces all week. Let me walk
you through some real examples that made
me go, "Okay, this is actually insane."
Last Tuesday, I gave it this prompt. I
need to organize a team off-site for 15
people in Portland next month. Find
three venue options with catering,
research team building activities, and
create a budget breakdown presentation.
And here's what happened. The agent
started browsing venue websites,
comparing pricing, checking availability
calendars, researching local activity
providers, and then it actually
generated a full PowerPoint presentation
with charts, venue photos, and a
detailed budget analysis. The whole
thing took about 20 minutes, and
honestly, it would have taken me half a
day to do manually. But what's really
clever is how it narrates everything
it's doing in real time. You can
literally watch it work and jump in with
corrections like actually make sure the
venues are dog friendly and it instantly
adapts without losing context. Another
example, I said research our top three
competitors pricing strategies and
create an executive summary. It went to
their websites, dug through pricing
pages, analyzed their positioning, and
generated a professional document with
citations and recommendations.
The level of analysis was genuinely
impressive.
Now, there are some limitations you
should know about. Pro users get 400
agent tasks per month, plus users get
40. And honestly, some tasks can take 15
to 30 minutes, so it's not instant
gratification. But for complex
multi-step workflows,
this is gamechanging. The setup is super
simple, too. You just click on agent in
the tools drop down and suddenly chat
GPT transforms from a chat interface
into this powerful workspace where it's
actually doing stuff in the background.
Codeex, the AI developer that never
sleeps. Now, if you're not a developer,
you might want to skip ahead, but trust
me, even if you don't code, this is
fascinating stuff. Codeex is basically
like having a senior developer on your
team who can work on your entire
codebase independently. I connected it
to one of my GitHub repos and gave it
this challenge.
Our website is loading slowly on mobile,
find the performance bottlenecks and fix
them.
What happened next blew my mind. It
analyzed the entire codebase, identified
that we were loading too many images up
front, implemented lazy loading,
optimized our CSS, and even wrote unit
tests to make sure nothing broke.
Then it created a pull request with a
detailed explanation of every change.
The technical specs here are nuts. It's
powered by this codeex 1 model with a
192,000 token context window. That means
it can hold your entire application in
memory while it works. Most AI coding
tools can only see a few files at a
time, but this thing understands your
whole project structure.
I also tested it with add a dark mode
toggle to our dashboard with smooth
animations and user preference
persistence.
It didn't just implement the feature, it
updated the design system, created the
toggle component, added the animation
CSS, set up local storage for
preferences, and even updated our
documentation.
The crazy part is it runs everything in
a sandboxed environment. So it can
actually test its changes and show you
proof that everything works before you
merge it into your main codebase. For
developers, this is honestly
revolutionary. It's like having a pair
programming partner who never gets
tired, never gets frustrated, and can
work on massive refactoring projects
while you sleep.
The model upgrade that changes
everything.
Okay, so beyond agent mode and codecs,
we also got some serious model upgrades
that are worth talking about. First up
is GPT4.1.
And the headline feature here is that
million token context window I
mentioned.
To put that in perspective, you could
literally paste in a 400page book and it
would remember every single detail
throughout your entire conversation.
I tested this by uploading our entire
company handbook and asking it to create
training materials. It pulled relevant
information from different sections and
created comprehensive guides that
actually made sense. The speed
improvements are also noticeable.
Tasks that used to take 30 to 40 seconds
are now happening in 15 to 20 seconds.
Not revolutionary, but when you're in a
flow state, those seconds add up. Then
there's 03 Pro, which is their new
reasoning model.
This thing is scary good at complex
problem solving. I gave it this
challenge.
Our customer churn rate increased 15%
last quarter.
Analyze our support tickets, user
feedback, and product usage data to
identify the root causes and recommend
specific solutions. It spent about 10
minutes really thinking through this,
analyzing patterns, cross-referencing
data points, and then delivered this
incredibly detailed analysis with
specific actionable recommendations.
The depth of reasoning was honestly
impressive. It connected dots that I
probably would have missed. The
trade-off is that 03 Pro is slower and
can't generate images, but for complex
analytical work, it's worth the wait.
Connectors.
Finally, AI that works with your real
data. This is where things get really
practical for most people. Connectors
let ChatGpt actually access your real
accounts. Gmail, Google Drive, Slack,
Calendar, GitHub, and more. The setup
process is straightforward. You go to
settings, click on connectors,
authenticate with each service, and
suddenly chat GPT can work with your
actual data instead of just
hypotheticals.
Here's a real example from last week. I
told it,
summarize the key decisions from our
leadership team meetings in November and
create action items for December. It
went into my calendar, found the meeting
recordings, analyzed the transcripts,
identified decisions and commitments,
and created a clean action item list
with owners and deadlines. For readonly
stuff, it accesses your accounts
directly after you authenticate. For
actions like sending emails, it uses
this secure browser takeover approach
where you log in normally and then it
continues using your session without
ever storing your passwords. There are
two modes worth knowing about. There's
the regular chat search mode for quick
questions like, "What did Sarah say
about the budget in her email
yesterday?"
And then there's deep research mode,
which is more intensive. It'll spend 10
to 15 minutes diving deep into multiple
sources to create comprehensive reports.
I have to say, having AI that can
actually work with my real data instead
of forcing me to copy paste everything
has been a huge workflow improvement.
Custom GPTs get a major upgrade. If
you've played with custom GPTs before,
you know they were pretty cool, but had
some limitations. Well, those
limitations just got blown up. The big
change is model selection. You can now
choose which specific model powers your
custom GPT. And this opens up some
really interesting possibilities. I
created three different custom GPTs to
show you what I mean. First, a social
media manager GPT using GPT 40 Mini for
speed. I gave it our brand guidelines
and content calendar and now I can just
say create this week's LinkedIn posts
and get five ready to publish posts in
about 30 seconds. Second, a financial
analyst GPT using 03 Pro for accuracy.
This one analyzes our quarterly reports,
industry trends and creates detailed
investment recommendations. It takes
longer but the depth of analysis is
worth it. Third, a creative director GPT
using the full GPT-4.1
for context. I uploaded our entire brand
archive and it can now create campaign
concepts that are perfectly aligned with
our brand history and evolution. The key
insight here is matching the model to
the task. You don't need a three pro to
write social media posts and you don't
want GPT40
mini handling complex financial
analysis. Building these is still super
straightforward. Define the role, choose
your model, upload any knowledge files,
set some behavioral guidelines, and
you're good to go. But now they're
actually optimized for their specific
purpose. Projects. Your AI command
center projects got some major upgrades,
too. And this feature is honestly
becoming the backbone of how I work with
AI. Now, think of projects as permanent
workspaces where chat GPT remembers
everything across multiple
conversations. I've got separate
projects for different clients. Our
product development content strategy,
each one maintains its own context and
memory. The new deep research
integration is fantastic. I can say
research sustainable packaging solutions
for our e-commerce client. Considering
the budget constraints we discussed last
month and it remembers our previous
conversations while pulling in new
research from the web. Voice mode in
projects is surprisingly useful. During
brainstorming sessions, I can just talk
through ideas while it takes notes and
asks clarifying questions.
It's like having a really good meeting
facilitator who never forgets anything.
The persistent memory across sessions is
the real game changer, though. I can
start a project on Monday, add some
files and have a few conversations, then
pick it up on Friday, and it remembers
every detail. No more let me catch you
up on what we discussed last time. I've
been using one project to plan our
annual company retreat. Over the past
few weeks, I've uploaded venue options,
budget spreadsheets, team feedback
surveys, and had dozens of conversations
about logistics. Now, when I ask for
updates or changes, it has the full
context of everything we've discussed.
Real world testing, what actually works.
All right, let's get real about what
this stuff actually does well and where
it still falls short because I've been
testing everything extensively. What
works brilliantly? Complex research
projects that would normally take hours.
Document creation where you need
professional formatting, data analysis
across multiple sources, routine
workflow automation, multi-step
planning, and coordination.
I had it plan my entire week last
Monday. It looked at my calendar,
prioritized my task list, researched
prep materials for meetings, and even
ordered lunch to be delivered during my
busiest day. Honestly, felt like having
a personal assistant. What's still
frustrating, sometimes it gets stuck in
loops on tricky websites.
The 20 to 30 minute task times can break
your flow, occasionally misinterprets
instructions, and goes down the wrong
path. Still needs human oversight for
anything important. The sweet spot seems
to be delegating those boring multi-step
tasks that you know how to do but just
don't want to spend time on. market
research, competitive analysis, data
compilation, presentation creation,
stuff that's important but not
particularly creative.
What this all means, here's the thing
that really strikes me about these
updates.
We're not just looking at incremental
improvements anymore. This feels like a
fundamental shift in what AI can do and
how we interact with it. 6 months ago,
AI was this thing you'd chat with to get
ideas or draft content. Now, it's
actually completing entire workflows
while you focus on the strategic stuff.
The productivity implications are
honestly staggering. For businesses, the
competitive advantage is going to come
from figuring out which processes to
automate and which ones need human
creativity. The companies that get this
balance right are going to move so much
faster than everyone else. But I think
the bigger picture is even more
interesting. We're moving toward this
world where AI isn't just a tool you use
occasionally. It's a collaborative
partner that's always working alongside
you. That changes everything about how
we think about work, productivity, and
honestly, what it means to be knowledge
workers. The early adopters who learn to
work effectively with these AI agents
are going to have a massive advantage.
This isn't about replacement. It's about
amplification. Look, I know this was a
lot to digest, and honestly, I'm still
discovering new capabilities every day.
These aren't just feature updates. They
represent a completely new category of
AI tools that can actually get stuff
done in the real world. If you're
already a ChatGpt subscriber, I
definitely recommend experimenting with
agent mode and the new models.
Start with smaller tasks to get a feel
for how they work, then gradually tackle
more complex workflows as you build
confidence.
If you're still on the fence about
upgrading, these features are probably
worth the subscription cost if you
regularly do research, analysis, or
content creation. The time savings alone
can be substantial. What I'm most
excited about is where this leads. If
this is what's possible now, imagine
what AI agents will be capable of in 6
months or a year. We're witnessing the
early stages of a fundamental shift in
how work gets done. What tasks would you
want to delegate to an AI agent first?
Are you excited about this stuff or
concerned about the implications? Let me
know in the comments. I read every
single one and they help shape what we
cover next. Don't forget to subscribe if
you want to stay on top of these AI
developments.
The landscape is changing so fast that
missing a few weeks can put you way
behind the curve. Thanks for watching
and I'll see you in the next one where
we dive even deeper into practical
applications and advanced workflows with
these new tools.