Latest / AI Fire Daily / #5 Eric: Stop Rebuilding Context and Stay Where the Work Lives

Transcript

#5 Eric: Stop Rebuilding Context and Stay Where the Work Lives · AI Fire Daily

0:00
You'll likely spend at least three hours this
0:03
week just copying and pasting text between different
0:06
tabs on your computer. Yeah, easily. It's the
0:09
modern knowledge worker's absolute worst nightmare.
0:11
Right, because you get this massive, incredibly
0:14
messy assignment. The decals are like... totally
0:17
buried in some endless email thread. Oh, and
0:20
the project assets are just scattered across
0:22
a completely chaotic Google Drive folder. Exactly.
0:26
And somehow, you are the one who needs to turn
0:29
all of this digital noise into, you know, a clean
0:32
document, a functioning project tracker, and
0:35
a polished slide deck for a leadership review.
0:37
Because the expectation is that you act as this
0:40
sort of human router. Like, your brain is just
0:42
being used to pull signal from the noise, format
0:45
it, and push it somewhere else. Which is incredibly
0:47
draining. And my biggest grievance with how people
0:49
use AI right now is that the standard quote unquote
0:52
AI workflow to solve this is just, I mean, it's
0:56
digital manual labor in disguise. It really is.
0:59
You're just substituting one form of busy work
1:01
for another. You find a file, you copy a giant
1:03
wall of text, you open up a separate AI chatbot
1:06
in another tab. and you painstakingly explain
1:09
all the context to it. Right. You wait for a
1:11
response, and then copy that response back into
1:14
your workspace. Do that 20 times a day, and you're
1:17
just as exhausted as before. Because bouncing
1:19
between tabs actually increases your cognitive
1:22
load, the technology is supposed to alleviate
1:24
that, not add to it. So our mission for this
1:27
deep dive is based on this fascinating new playbook
1:31
for using Gemini embedded directly inside Google
1:33
Workspace. Yeah. It's a total paradigm shift.
1:36
how to stop this endless copy -pasting cycle.
1:40
We're looking at turning a chaotic web of files
1:43
into a connected workflow. like without ever
1:45
leaving the ecosystem where your work already
1:47
lives. And if we distill this entire playbook
1:50
down to one golden rule, it is this embedded
1:53
AI saves the most time when it reduces context
1:56
transfer, not when it simply generates more content.
2:00
Okay, wait, context transfer. What exactly do
2:02
you mean by that? Well, generating text is the
2:04
easy part. Moving context from your brain or
2:06
from one app to another, that's where the friction
2:09
lies. Oh, I see. So the real enemy here isn't
2:11
a lack of ideas. It's what developers call context
2:15
switching. Exactly. The brain drain of having
2:17
to re -explain the background of a project every
2:20
single time you open a new tab or start a new
2:22
task. Is that the core of context transfer? That
2:25
captures it perfectly. It's the friction of repeatedly
2:29
establishing the baseline facts. A major mistake
2:31
people make is assuming context just, you know,
2:34
automatically carries across every app they use.
2:36
And traditionally, it really doesn't. No, it
2:38
doesn't at all. But with Gemini inside Workspace,
2:41
five core apps, so Gmail, Drive, Docs, Sheets,
2:46
and Slides are now designed to support a single
2:49
connected workflow. The practical takeaway here
2:52
is to stay in workspace when both your source
2:54
material and your final output already live there.
2:57
Stay where the work lives. I like that. So to
3:00
fix this copy -paste nightmare, we kind of have
3:03
to follow the life cycle of a project. And the
3:05
mess almost always originates in the inbox. Yeah,
3:08
let's look at Gmail because that is where the
3:10
grenade usually gets dropped, right? Like a manager
3:12
hands over a broken project or a client suddenly
3:16
pivots their strategy halfway through a massive
3:19
ongoing threat. And the real task you need to
3:21
accomplish is rarely sitting neatly at the top
3:24
of that email. Oh, never. Right. It's usually
3:26
buried under like four different greetings, three
3:29
layers of quoted replies. And side conversations
3:31
about someone's weekend. Exactly. It requires
3:34
an archaeological dig just to figure out what
3:36
you're actually supposed to do. But using Gemini
3:39
in Gmail isn't just about asking it to draft
3:43
a polite reply, it's about having the AI summarize
3:46
these convoluted threads and extract the specific
3:49
actionable details. Because the underlying mechanism
3:52
here is semantic understanding. The AI isn't
3:55
just looking for keywords like deadline or budget.
3:58
It's doing more than just a word search. Way
4:00
more. It's building a relationship map of the
4:03
conversation to figure out who owes what to whom.
4:06
Gmail's role fundamentally shifts from a communication
4:09
tool to a translation tool. It turns unstructured,
4:12
messy human conversation into a concrete starting
4:15
point. Exactly. Okay, so we have our concrete
4:18
starting point from the inbox. We know what the
4:20
boss wants. But a project doesn't live in an
4:23
email. It lives in a shared folder. Right. The
4:26
dreaded shared drive. Yeah. We have to assess
4:28
the actual damage across the company's files.
4:31
That brings us to Google Drive, which sort of
4:34
serves as the project brain. And I have to push
4:36
back a little here because Drive is usually where
4:39
a project looks incredibly organized right up
4:41
until you actually need to understand it. Oh,
4:43
totally. It's just a folder filled with 10 documents,
4:46
all with variations of the name brief final,
4:49
brief final v2. And final seven re - Yeah, the
4:52
real final file name is a universal constant.
4:55
So if you're listening to this and thinking,
4:56
you know, my company's drive folders are way
4:59
too chaotic for an AI to make sense of how does
5:02
Gemini actually parse that mess, let's use the
5:05
specific scenario from the source material to
5:07
ground this. Let's do it. Let's say you've just
5:09
taken over the Northstar Analytics website relaunch.
5:11
A classic terrible handover scenario. It is a
5:15
complete disaster. The handover notes are in
5:18
that Gmail thread we just summarized. The actual
5:21
project history is scattered across nine different
5:24
drive files with terrible names. The pricing
5:27
is blocked. It's currently forecasting $1 ,000
5:30
over budget. And you have an executive review
5:32
coming up fast. Right. So in a normal world,
5:35
how much time are you spending just opening files
5:38
to figure out what is happening? In a traditional
5:40
workflow. You're spending your next three hours
5:42
with a legal pad, opening every single document,
5:46
skimming for updates, and trying to reconcile
5:48
conflicting information manually. But with this
5:50
playbook, you take that Gmail summary, you paste
5:53
it into the Gemini panel in Drive, and things
5:56
change. Drive stops being a static storage cabinet.
6:00
Yeah, the AI looks past the chaotic file names
6:02
and analyzes the actual contents. It synthesizes
6:05
those nine scattered files. Which is wild. And
6:07
it goes beyond a vague overview. It extracts
6:10
specific operational facts. Like it tells you
6:12
that exactly six out of 18 tasks are complete.
6:15
It spots the thousand dollar budget variance.
6:18
It even flags that one specific SEO task still
6:22
doesn't have an owner. So Drive isn't just a
6:24
storage unit. It's acting like a like a triage
6:27
nurse in an emergency room. It takes one look
6:30
at the Bleeding North Star project, figures out
6:32
what is critically broken, and tells the other
6:35
departments exactly what tools to prep. That's
6:37
a great analogy. Because the most crucial detail
6:40
I found in this section is that Drive actually
6:43
generates the specific prompts you will use in
6:45
the other Workspace apps. Yes. It hands you the
6:48
exact instructions to feed into Docs, Sheets,
6:51
and Slides. And more importantly, it tells you
6:53
which specific files to attach to those prompts.
6:56
So you aren't just blindly throwing every project
6:58
file into an AI and hoping it understands. Exactly.
7:00
You are curating the exact context package. That
7:03
is a massive reduction in cognitive load. Oh,
7:05
and a quick detour. For our mobile listeners,
7:08
there was a fascinating update mentioned in the
7:11
sources. Oh, the document scanner. Yeah. Drive's
7:13
built -in document scanner on Android and iOS
7:16
can now scan multiple physical pages and automatically
7:19
separate them into clean, digitized documents.
7:22
That's huge. Right. So even if part of this messy
7:24
project handover included literal paper notes
7:27
on someone's desk, you can pull them into this
7:30
connected digital workflow instantly. Bridging
7:32
the gap between analog and digital context is
7:35
just vital these days. So, our triage nurse has
7:38
diagnosed the North Star Analytics mess, highlighted
7:41
the budget issues and the unassigned tasks, and
7:43
handed you the prompts. But a diagnosis doesn't
7:46
cure the patient. We need a recovery strategy
7:48
that other humans can actually read and follow.
7:51
Scattered information has become a single cohesive
7:54
plan. Which takes us into Google Docs. Yes, but
7:58
I have a major question about this step. If I
8:00
go into docs and ask an AI to write a recovery
8:03
plan, aren't I just going to get a generic robotic
8:05
outline? Like, step one, assess the situation.
8:09
Step two, implement solutions. That doesn't help
8:11
anyone. And that is the exact trap people fall
8:14
into when they use standalone AI tools without
8:16
injecting context. They get boilerplate text.
8:18
Right. But the magic formula outlined in the
8:20
playbook prevents this. It requires three specific
8:23
inputs, a design reference, your selected sources,
8:26
and the drive -generated prompt. Okay, let's
8:28
break the mechanics of that formula down. How
8:31
does the design reference actually work? So,
8:34
when you open a blank doc, you use a feature
8:36
called match doc format. You select a previous
8:39
business report you wrote, one that has your
8:40
company's preferred layout, font hierarchy, and
8:43
professional tone. Oh, so it mimics your style.
8:46
Exactly. The AI analyzes the structural parameters
8:49
of that document. That's your design reference.
8:51
Next, you attach only the specific files the
8:54
Drive triage process told you to attach. That
8:57
is your source material. Got it. And finally,
9:00
you paste in the highly specific prompt Drive
9:02
wrote for you. So you aren't starting from a
9:05
blank page or some generic internet template.
9:08
Gemini is extracting your actual facts, like
9:11
the specific $1 ,000 variance, the real milestones,
9:14
and embedding them into a highly specific recovery
9:16
plan formatted exactly the way your company expects
9:19
it to look. It synthesizes the mess into a polished,
9:22
strategic document covering resource constraints,
9:25
owners, and decisions needed without sounding
9:27
like a generic chatbot. Which is fantastic for
9:30
strategy. But, you know, in my experience, a
9:32
beautifully written narrative document? doesn't
9:35
actually hold anyone accountable. Nope. People
9:37
read paragraphs, nod their heads, and then miss
9:40
their deadlines. We need a grid. We need to know
9:44
who's doing what by when. The operational reality
9:47
of any project. Exactly. And the source material
9:49
has a brilliant line about this transition into
9:52
Google Sheets. It says, if the tracker takes
9:55
longer to maintain than the project, you have
9:57
gone too far. Oh, we have all built those monstrous
10:00
color -coded spreadsheets that require a full
10:03
-time job just to update. Yes. So the goal is
10:06
to build a tracker that serves the project rather
10:09
than forcing the project to serve the tracker.
10:11
Absolutely. So how does the AI translate paragraphs
10:14
of strategy into an operational grid? Like you
10:18
repeat the handoff process right you import the
10:21
recommended sources into sheets and you paste
10:23
the prompt drive gave you right But Gemini isn't
10:26
just pasting text into a cell it actually structures
10:29
the tracker wait like building the columns Yeah,
10:32
it creates the appropriate columns for tasks,
10:34
owners, due dates, and status. It handles spreadsheet
10:38
-specific actions. It understands that status
10:41
shouldn't just be open text. It needs to be a
10:43
drop -down menu. And it writes formulas. That
10:46
blew my mind. It replaces static numbers with
10:48
functional logic. It's pretty incredible. Looking
10:50
back at our Northstar project, we had that one
10:52
unassigned SEO task. Gemini sets up the KPI section
10:57
of this sheet so that if you assign an owner
10:59
to that SEO task, the unassigned count automatically
11:03
drops from 1 to 0. And the completion rate automatically
11:06
updates when a status changes to done. You are
11:09
generating a functional tool based on the linguistic
11:12
context of your emails. How does a language model
11:14
even do that? Well, under the hood, the AI is
11:17
translating natural language instructions into
11:19
the underlying coder functions that power the
11:22
spreadsheet. It maps the concept of, say, tracking
11:24
completion to the specific formula required to
11:28
calculate a percentage based on drop -down variables.
11:30
So it bridges the gap between how humans speak
11:33
and how spreadsheets compute. Exactly, which
11:35
is a huge leap from just generating text. Okay,
11:38
so the team now has a functional tracker. They
11:41
know what to do. But eventually, you have to
11:44
present this operational reality to leadership.
11:47
And let's be honest, executives do not have the
11:50
time or the desire to decipher a dense spreadsheet
11:53
or read a 12 -page recovery document. They want
11:56
the five -minute version. The art of compression.
11:58
We have to move into Google Slides. We need to
12:00
take all this data and make it visual. Right.
12:02
The final deck for this executive review needs
12:04
to focus strictly on the new timeline, current
12:07
status, major blockers, resource constraints,
12:10
and the specific decisions required to unblock
12:12
that pricing issue. But compression is arguably
12:15
a much harder cognitive task for an AI than generation,
12:18
isn't it? Yeah. Docs holds the full narrative.
12:20
Sheets holds the granular detail. But Slides
12:23
requires ruthless editing. You are asking the
12:25
model to decide what information is safe to discard.
12:28
It is a totally different challenge. You give
12:30
slides the specific context package and a visual
12:33
reference deck to work from. But, and this is
12:35
important, the playbook comes with a pretty severe
12:37
warning about AI and presentations. Yes. It raises
12:40
a crucial point about how large language models
12:42
operate. AI has never met a paragraph it didn't
12:46
want to put on a slide. They are predictive text
12:48
engines. Their inherent design is to generate
12:50
more tokens to keep explaining. They basically
12:53
default to verbosity. They lack spatial awareness.
12:57
Like an AI doesn't inherently understand that
12:59
a human sitting 20 feet away from a screen simply
13:03
cannot read a wall of 12 point font. Precisely.
13:06
The AI processes tokens sequentially, not visually.
13:11
So while Gemini can do the heavy lifting of the
13:13
first pass, like structuring the presentation
13:16
flow around the core facts. Visual review by
13:19
a human is absolutely mandatory here. You have
13:21
to go in and enforce visual hierarchy. You must
13:23
ensure it hasn't crammed a massive paragraph
13:25
where a three -word bullet point should be. You
13:27
have to act as the editor. Okay, let's look at
13:29
the full picture. We took a chaotic email thread,
13:31
diagnosed it with our drive triage nurse, built
13:34
a strategic recovery plan in docs, created an
13:36
automated tracker in sheets, and compressed the
13:39
entire thing into an executive presentation in
13:41
slides. But now the loop has to be closed. Right.
13:45
The deliverables are useless if they just sit
13:47
on your hard drive. We go back to the origin
13:49
point. Gmail. We draft a stakeholder message.
13:53
We tell Anna, our executive sponsor, that the
13:55
plan is ready for review. And we ping finance
13:58
to unblock that pricing issue. And according
14:00
to the playbook, this massive multi -app recovery
14:03
task can be handled from start to finish. in
14:06
under 30 minutes. Which is staggering when you
14:08
evaluate the manual alternative of doing all
14:11
that context transfer yourself. It changes the
14:13
math of productivity entirely. It really does.
14:16
But you know, before we assume AI can flawlessly
14:19
execute our jobs, and we can all just go take
14:21
a nap, we have to examine where the system breaks
14:24
down. Knowing the limitations of a tool is arguably
14:26
more important than knowing its features. Always.
14:28
If embedded Gemini inside these workspace apps
14:31
is this interconnected and powerful, why would
14:34
a user ever open the standalone Gemini app again?
14:37
Like, what is the difference? That is a critical
14:39
distinction for anyone using these tools. Embedded
14:41
Gemini, the sidebars inside docs or sheets, is
14:44
fundamentally tied to the specific job of the
14:46
application you are currently in. So it's tuned
14:49
for context -heavy execution of a known task.
14:52
Formatting a tracker, summarizing a thread, drafting
14:55
an outline. Exactly. It is focused on the files
14:58
directly in front of it. But if you need broad
15:01
complex reasoning, deep market research, coding
15:04
assistance, or flexible open -ended brainstorming
15:08
that isn't tethered to your specific internal
15:10
documents. Then you still need the standalone
15:12
app. Yes. The embedded version is an execution
15:15
engine. The standalone version is a wide open
15:17
thinking partner. Their underlying parameters
15:20
are tuned for different cognitive tasks. That
15:22
makes perfect sense. And the source material
15:25
highlights some critical blind spots when relying
15:27
on that execution engine. What happens when the
15:29
AI simply gets the facts wrong? And this is where
15:32
the human element remains non -negotiable. Gemini
15:35
can and will make what I call confident mistakes.
15:38
I love that phrase, confident mistakes. It sounds
15:40
so authoritative and sure of itself while being
15:43
completely, utterly wrong. Why does that happen
15:46
even when it's reading your own company files?
15:49
Because LLMs are not traditional databases. They
15:52
do not look up a fact in a rigid table. They
15:55
predict the next most likely word based on patterns.
16:00
So it might make an unsupported assumption about
16:02
a deadline based on standard business practices
16:04
rather than the specific text in your email.
16:08
It might assign the wrong owner to a task or
16:10
misinterpret a budget variance as a surplus instead
16:13
of a deficit. You must meticulously double -check
16:16
anything that affects budgets, timelines, or
16:19
responsibilities. Because if you send Anna a
16:21
slide deck claiming the North Star budget is
16:23
perfectly fine when it's actually $1 ,000 in
16:25
the red, you cannot just tell your boss, oh,
16:28
the AI sidebar made a mistake. The accountability
16:31
still rests entirely on you. Exactly. Another
16:34
significant blind spot occurs with scale. Context
16:37
windows, which is the amount of text an AI can
16:40
hold in its short -term memory, are expanding.
16:42
But they aren't infinite. If your source set
16:45
in a drive folder becomes too massive, the AI
16:48
might miss a crucial file entirely. Or it might
16:52
focus heavily on an older update from last month
16:55
and ignore the newest version from yesterday.
16:57
You have to verify that it is working from the
17:00
most recent, accurate reality. It also fundamentally
17:03
doesn't own the workflow. It's an assistant,
17:06
not an autonomous agent. Like, it isn't going
17:08
to look at your calendar, realize it's Friday
17:10
at 4 PM, and automatically email your boss a
17:14
project update. No, you have to initiate every
17:16
single step. You click to summarize the inbox,
17:19
you trigger the drive triage, you open the dock,
17:22
you paste the prompt, you are still the driver
17:24
of the car. It removes the friction of context
17:27
transfer, but it does not remove the responsibility
17:29
of project management. You have to guide the
17:31
tool at every juncture. Before we bring this
17:34
all together, I want to share a really awesome
17:36
bonus fact from the material, specifically for
17:38
any students listening. Oh yeah, this is a great
17:40
update. Google just announced that eligible U
17:42
.S. college students can get Google AI Pro for
17:45
free for an entire year. And for students in
17:48
over 140 other countries, you can get Google
17:51
AI Plus for free for a year. Giving students
17:54
early access to these tools is a massive advantage.
17:57
Learning how to engineer prompts and manage complex
17:59
information workloads before entering the workforce,
18:02
that's going to be a critical career differentiator.
18:05
Learning how to guide the AI rather than just
18:07
asking it to do your homework is a skill that
18:10
will pay massive dividends. Okay, let's bring
18:13
it all home. The core takeaway from everything
18:16
we've explored today is this. Stop treating your
18:18
workspace apps as separate, isolated silos. They
18:21
aren't different rooms you walk between. They're
18:23
simply different lenses through which to view
18:26
the exact same project data. Allow Gemini to
18:29
handle the repetitive, exhausting work of transforming
18:31
information from a conversation into a plan,
18:34
a plan into a tracker, and a tracker into a presentation.
18:37
But keep your human hands firmly on the wheel
18:39
when it comes to verifying the critical facts,
18:42
making the final strategic decisions, and managing
18:44
the human relationships with your stakeholders.
18:46
Let the AI suffer through the heavy lifting of
18:48
context transfer. so you can preserve your cognitive
18:51
energy for actual strategy. It shifts your role
18:54
from data entry to data evaluation. Analyzing
18:59
this entire workflow raises a really important
19:02
question, something I think is worth mulling
19:04
over long after this deep dive ends. I am always
19:07
here for a good, provocative thought. What are
19:09
you thinking? Well, we have established that
19:11
AI is now perfectly capable of doing this busy
19:13
work. It can translate a messy email handover
19:16
into a beautiful spreadsheet tracker and a consent
19:19
slide deck in half an hour. And it is a massive
19:21
win for immediate productivity. Unquestionably,
19:24
nobody wants to spend three hours formatting
19:26
a grid. Right. True. But historically, doing
19:29
that exact kind of repetitive information formatting
19:32
busy work is how junior employees and entry level
19:36
analysts actually learn the inner workings of
19:38
a business. Oh, wow. You learn what makes a project
19:41
tick. You learn the company's priorities by being
19:44
forced to manually synthesize messy details into
19:46
a slide deck week after week. You absorb the
19:49
context by suffering through the transfer. That
19:52
is a fascinating paradox. We are automating the
19:55
exact friction that traditionally builds expertise.
19:58
Precisely. If embedded AI now takes over all
20:01
that context transfer and synthesis, what happens
20:04
to those traditional entry level learning experiences?
20:07
If Gemini is instantly spotting that the SEO
20:10
task is unassigned or the budget is over by a
20:13
thousand dollars, how will the next generation
20:15
of workers develop the intuition to spot a failing
20:18
project on their own? Are we removing the training
20:21
wheels before the rider has learned how to balance?
20:24
It really makes you wonder how the analysts of
20:26
tomorrow are going to earn their stripes and
20:28
develop that gut instinct for business logic.
20:30
You start out just wanting to clean up a messy
20:32
project folder, and you end up questioning the
20:34
future of how human beings learn their trades.
20:36
It's a lot to think about. It really is. We started
20:39
today talking about that awful feeling of being
20:41
handed a chaotic web of scattered files and endless
20:45
emails, the pure manual labor of copying and
20:47
pasting just to figure out where to begin. Hopefully
20:50
you now have a playbook to navigate that muddy
20:52
water, to let the tools do the synthesis so you
20:55
can focus on the strategy. Thank you for joining
20:57
us on this deep dive. We highly encourage you
21:00
to try building your own connected workflow today.
21:03
Go see what you can build when you stop fighting
21:04
the context transfer. Catch you next time.