Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
(00:00):
Welcome to the AI Breakdown.
(00:01):
I'm Andy, and today we're diving into something truly fascinating about the future of work and technology.
We're talking about AI agents, what they are, how they work, why they're gaining traction, and how you can use them.
I'll be sharing real world examples, practical advice, and an look at how these agents are transforming everything from your cluttered inbox to entire operations.
(00:27):
And I'll give you my honest take on chat.
GT's newly released agent mode that's been making waves these past few weeks.
7
00:00:40,781.781081849 --> 00:00:42,281.781081849
Let's start with the basics.
8
00:00:42,581.781081849 --> 00:00:52,631.781081849
What exactly is an AI agent picture? A new intern, Barry, you tell Barry Sort my emails, book my meetings, draft a report, and arrange my trip to London.
9
00:00:53,171.781081849 --> 00:00:59,621.781081849
Barry nods tackles the list, possibly asks a few clarifying questions and adapts when he hits a roadblock.
10
00:01:00,266.781081849 --> 00:01:02,846.781081849
That's precisely what an AI agent is.
11
00:01:03,116.781081849 --> 00:01:10,376.781081849
Not just a rigid script following orders, but a digital helper who takes your goal, figures out the steps needed and adjusts course.
12
00:01:10,376.781081849 --> 00:01:11,516.781081849
When things get tricky.
13
00:01:12,356.781081849 --> 00:01:20,546.781081849
These agents excel at breaking down complex goals, making decisions, remembering important context, and using various tools to get the job done.
14
00:01:21,176.781081849 --> 00:01:26,31.781081849
Unlike traditional software or those frustrating chatbots that just answer once and disappear.
15
00:01:26,741.781081849 --> 00:01:33,491.781081849
Agents stick with the task constantly evaluating, updating their approach and working until they finish the job.
16
00:01:33,821.781081849 --> 00:01:36,221.781081849
Now let's dig under the hood and look at the machinery.
17
00:01:36,581.781081849 --> 00:01:45,671.78108185
The brain of a modern AI agent is usually a large language model, think GPT-4 O or something similar that powers chat, GPT.
18
00:01:46,361.78108185 --> 00:01:51,281.78108185
This allows the agent to understand your request reason about it and determine what comes next.
19
00:01:51,611.78108185 --> 00:01:57,101.78108185
But the real magic lies and how agents remember, use tools and operate independently.
20
00:01:57,701.78108185 --> 00:02:01,991.78108185
Memory is crucial, both short and long term agents track progress.
21
00:02:02,21.78108185 --> 00:02:06,581.78108185
What tasks are done, your instructions even details from past conversations.
22
00:02:07,211.78108185 --> 00:02:08,261.78108185
Then there's tool use.
23
00:02:08,651.78108185 --> 00:02:10,361.78108185
Agents can plug into web searches.
24
00:02:10,541.78108185 --> 00:02:12,911.78108185
They can research query databases.
25
00:02:13,301.78108185 --> 00:02:25,721.78108185
Interact with email platforms and even run bits of code or actions in other apps, expanding their capabilities well beyond simple conversations into genuine digital colleagues that can take meaningful action.
26
00:02:26,261.78108185 --> 00:02:28,361.78108185
The glue holding this together is autonomy.
27
00:02:28,751.78108185 --> 00:02:31,661.78108185
Agents plan, act, check results, and adjust.
28
00:02:32,141.78108185 --> 00:02:38,561.78108185
Sometimes they pull from existing knowledge, sometimes they work as teams passing tasks between specialists.
29
00:02:38,996.78108185 --> 00:02:47,306.78108185
It all comes together to create a workflow that mirrors how we humans operate, sensing our environment, planning our next move, taking an action.
30
00:02:47,696.78108185 --> 00:02:52,556.78108185
Learning from results, and repeating the end product isn't just clever software.
31
00:02:52,856.78108185 --> 00:02:56,906.78108185
It's a genuine digital colleague that can handle meaningful work independently.
32
00:02:57,656.78108185 --> 00:03:04,16.78108185
Think of it as crossing the line from smart tool to team member, freeing you up to focus on what humans do best.
33
00:03:04,636.78108185 --> 00:03:12,826.78108185
Now let's explore some real world AI agents in action, starting with chat GT's, exciting new agent mode, which was released just a few weeks ago.
34
00:03:13,441.78108185 --> 00:03:27,421.78108185
Open AI's aim here is to combine the best of their offering chat, GPT, deep Research and operator, to seamlessly blend conversation, analysis and real world action into a single, unified and smooth experience.
35
00:03:27,781.78108185 --> 00:03:33,436.78108185
How does this all work? Chat GPTs Agent Mode uses a secure virtual computer.
36
00:03:33,826.78108185 --> 00:03:39,436.78108185
It can browse the web, click through pages, interact with your files, and connect to apps you use.
37
00:03:39,466.78108185 --> 00:03:43,666.78108185
Think Gmail, Google Calendar, or even GitHub if you are a techie.
38
00:03:43,966.78108185 --> 00:03:47,596.78108185
And don't worry, it asks for your permission before doing anything important.
39
00:03:48,331.78108185 --> 00:03:49,141.78108185
The big idea.
40
00:03:49,441.78108185 --> 00:04:08,911.78108185
Instead of you copy and paste in answers or manually juggling steps, chat, GPT can handle it hands free, showing you exactly what it's doing all along, like researching and booking your next holiday or sorting through your overflowing email inbox, finding important messages, drafting replies, and even scheduling meetings for you.
41
00:04:09,871.78108185 --> 00:04:16,771.78108185
All of this happens while you stay in control, freeing up your time and letting the AI do the heavy lifting behind the scenes.
42
00:04:17,206.78108185 --> 00:04:30,946.78108185
But how does it perform in the real world? I've been putting chat GPTs agent mode through its paces over the past few weeks, and I have to say I'm really disappointed, which is big coming from me as I'm typically a huge fan of their new releases.
43
00:04:31,576.78108185 --> 00:04:33,376.78108185
Let me share a real world test.
44
00:04:33,646.78108185 --> 00:04:36,166.78108185
I tried to get it to help book a hotel for my family.
45
00:04:36,436.78108185 --> 00:04:37,816.78108185
Visiting from overseas.
46
00:04:38,236.78108185 --> 00:04:41,116.78108185
I activated the agent mold and gave it these instructions.
47
00:04:41,596.78108185 --> 00:04:46,276.78108185
My dad, stepmom and two brothers, both adults are in Newcastle from Friday.
48
00:04:46,786.78108185 --> 00:04:48,286.78108185
I need to help them book a hotel.
49
00:04:48,736.78108185 --> 00:04:53,956.78108185
My dad, one brother and stepmom are leaving on Wednesday, my other brother on Tuesday.
50
00:04:54,226.78108185 --> 00:04:57,796.78108185
I want a hotel on the river with decent reviews, at least eight out of 10.
51
00:04:58,726.78108185 --> 00:05:00,346.78108185
Initially things looked promising.
52
00:05:00,706.78108185 --> 00:05:03,406.78108185
The agent correctly identified the dates and location.
53
00:05:03,796.78108185 --> 00:05:12,586.78108185
No small feat considering there are several new castles around the world and suggested some decent hotels on the river and included reviews and average ratings.
54
00:05:13,156.78108185 --> 00:05:15,466.78108185
It even asks the following, clarifying questions.
55
00:05:15,946.78108185 --> 00:05:24,856.78108185
Let me know which hotel you prefer and whether you want two rooms for all five nights, or need one room only for the last night since one brother leaves Tuesday.
56
00:05:25,366.78108185 --> 00:05:32,146.78108185
Once you decide I can proceed with the booking, this happened all within 12 minutes, but here's where things went pear shaped.
57
00:05:32,506.78108185 --> 00:05:35,566.78108185
I responded with my preferences and asked her to book my chosen hotel.
58
00:05:36,856.78108185 --> 00:05:47,776.78108185
But because I didn't explicitly reengage agent mode, it suddenly reverted to Standard Chat GPT Assistant mode and merely provided more information rather than actually helping with the booking.
59
00:05:48,256.78108185 --> 00:05:52,996.78108185
Not exactly a seamless user experience, though I imagine they'll sort this out eventually.
60
00:05:53,356.78108185 --> 00:05:58,936.78108185
So I deliberately reengaged agent mode, repeated my answers, and asked it to continue with the booking.
61
00:05:59,566.78108185 --> 00:06:03,226.78108185
Seven minutes later, it asked me to consent to cookies on the hotel's website.
62
00:06:03,511.78108185 --> 00:06:07,861.78108185
Which I did, but then it simply failed with this totally unhelpful message.
63
00:06:08,341.78108185 --> 00:06:14,311.78108185
It looks like the computer tool has been disabled, so I can't complete the booking or check live prices right now.
64
00:06:15,91.78108185 --> 00:06:17,491.78108185
Not exactly the digital revolution we were promised.
65
00:06:17,491.78108185 --> 00:06:17,911.78108185
Is it.
66
00:06:18,436.78108185 --> 00:06:24,136.78108185
I've also tested it with project work, hoping it might shine beyond what normal deep research can achieve.
67
00:06:24,496.78108185 --> 00:06:26,86.78108185
But again, I wasn't impressed.
68
00:06:26,536.78108185 --> 00:06:38,26.78108185
I actually got the same quality results more quickly through standard deep research itself, open AI clearly has quite a way to go before this becomes genuinely useful for everyday practical tasks.
69
00:06:38,626.78108185 --> 00:06:41,116.78108185
That said, I remain cautiously optimistic.
70
00:06:41,116.78108185 --> 00:06:42,616.78108185
They'll get it right eventually.
71
00:06:43,306.78108185 --> 00:06:45,946.78108185
So you've just heard my tail slash whinge of chat.
72
00:06:45,946.78108185 --> 00:06:50,206.78108185
GT's, new agent mode, failing on a fairly simple hotel booking.
73
00:06:50,656.78108185 --> 00:06:55,606.78108185
Now let's balance the scales with a set of AI agents that are genuinely delivering results right now.
74
00:06:56,66.78108185 --> 00:07:04,166.78108185
Starting with Gong's revenue, AI platform, the brilliant sales companion that's transforming customer conversations into revenue, all behind the scenes.
75
00:07:04,345.35251042 --> 00:07:12,775.35251042
Imagine every customer call, email, and zoom meeting automatically transcribed, analyzed, and distilled into crystal clear next steps.
76
00:07:13,150.35251042 --> 00:07:17,320.35251042
Without a human frantically scribbling notes, that's gong's specialty.
77
00:07:17,680.35251042 --> 00:07:24,460.35251042
The AI agents sit in the background, digest billions of data points, then surface actions that actually move deals forward.
78
00:07:25,330.35251042 --> 00:07:27,940.35251042
One of their standout features is smart trackers.
79
00:07:28,360.35251042 --> 00:07:33,220.35251042
Imagine AI agents that you can train to flag exactly when a competitor's name pops up.
80
00:07:33,670.35251042 --> 00:07:42,40.35251042
When a prospect raises pricing concerns or when there's that golden buying signal, and the results speak for themselves, deals where teams switched on.
81
00:07:42,40.35251042 --> 00:07:46,750.35251042
These smart trackers close a whopping 35% more often than those without.
82
00:07:47,480.35251042 --> 00:07:56,125.35251042
Granted, this is gong marking their own homework, but it's based on over a million analyzed opportunities and across 1,400 plus sales orgs.
83
00:07:56,542.59740838 --> 00:08:03,712.59740838
Then there's their brilliantly named Ask Anything feature, a conversational agent that transforms how sales teams get insights.
84
00:08:04,312.59740838 --> 00:08:05,842.59740838
Picture a rep simply typing.
85
00:08:06,217.59740838 --> 00:08:18,127.59740838
Wise, Acme Corporation dragging their feet and instantly receiving a comprehensive answer pulled from all previous calls, emails, and interactions, no digging through notes or transcripts.
86
00:08:18,607.59740838 --> 00:08:23,977.59740838
Teams leveraging this natural language query tool saw their win rates jump by 26%.
87
00:08:24,547.59740838 --> 00:08:26,377.59740838
And that's just the tip of the iceberg.
88
00:08:26,797.59740838 --> 00:08:32,257.59740838
Gong has a treasure trove of additional features that are revolutionizing sales teams worldwide.
89
00:08:32,524.8696367 --> 00:08:38,254.8696367
As you may know from past episodes, we use Gong at Notify Technology, and I've seen its benefits firsthand.
90
00:08:38,734.8696367 --> 00:08:45,394.8696367
On the downside, it's slow to transcribe calls and meet and notes, which isn't great in a fast-paced sales environment.
91
00:08:45,754.8696367 --> 00:08:51,364.8696367
And with Gong being positioned as a premium platform, it ain't cheap, so it can be cost prohibitive.
92
00:08:51,618.13494282 --> 00:08:56,268.13494282
So what other The innovative AI agents are making waves right now in customer support.
93
00:08:56,388.13494282 --> 00:09:08,988.13494282
Agents like Aada are designed to handle entire customer interactions autonomously, fetching account information, troubleshooting issues, resolving routine problems, and escalate into humans only when needed.
94
00:09:09,318.13494282 --> 00:09:15,768.13494282
Each AI agent deployed by ADA can perform the equivalent work of 10 plus human agents for less than the cost of one.
95
00:09:16,368.13494282 --> 00:09:21,258.13494282
Leading to major efficiency gains and substantial cost savings for organizations.
96
00:09:21,888.13494282 --> 00:09:32,418.13494282
This highlights the transformative potential of advanced AI agents and customer support, dramatically reducing workload for human teams, and speeding up help for customers in marketing.
97
00:09:32,478.13494282 --> 00:09:44,508.13494282
AI platforms like Jasper use advanced natural language processing to generate high quality branded content, such as blogs, emails, ad copy, and social media posts efficiently and at scale.
98
00:09:45,228.13494282 --> 00:10:06,648.13494282
Jasper itself stands out by enabling seamless integration of brand voice, supporting over 100,000 users and 850 plus enterprises as of 2024, with features like customizable templates, AI powered SEO optimization and autonomous marketing agents that manage campaign tasks and personalization.
99
00:10:07,198.13494282 --> 00:10:13,348.13494282
76% of software developers are using or plan to use AI coding assistance in their daily workflow.
100
00:10:13,828.13494282 --> 00:10:19,648.13494282
Things like GitHub's, copilot, or cursor to write test, and even deploy code with minimal human help.
101
00:10:20,398.13494282 --> 00:10:24,209.13494282
All these tools are introducing agents that move the interaction from assist to do.
102
00:10:24,802.42065711 --> 00:10:35,32.42065711
And non-techies now have access to vibe coding tools like Lovable and Bolt New to build software through conversation, literally chatting to an agent in plain English.
103
00:10:35,332.42065711 --> 00:10:51,842.42065711
Whilst it does all the heavy lifting and coding, We're seeing big companies like Pepsi and LinkedIn using AI agents heavily for things like recruitment, hr, or insurance underwriting, where agents handle screening at scale, analyze trends, and keep business flowing.
104
00:10:52,93.03290201 --> 00:10:58,663.03290201
So agents work across personal productivity, customer service, sales, software, hr, and beyond.
105
00:11:00,509.72948838 --> 00:11:06,509.72948838
Let's be brutally honest about where AI agents truly shine and where they stumble their strengths.
106
00:11:06,869.72948838 --> 00:11:13,469.72948838
They're the ultimate employees working 24 7 without complaints handling those mind numbing, repetitive tasks.
107
00:11:13,964.72948838 --> 00:11:20,84.72948838
Maintaining consistency and seamlessly connecting with your tools and systems to perform tasks at scale.
108
00:11:20,444.72948838 --> 00:11:33,974.72948838
But the flip side, well, as you heard from my chat, GPT Hotel Booking Fiasco, these digital assistance can go properly rogue, sometimes in hilarious ways, sometimes in ways that make you question your life choices.
109
00:11:34,484.72948838 --> 00:11:41,684.72948838
I've seen agents get trapped in bizarre loops, repeating the same tasks, endlessly like a technological version of Groundhog Day.
110
00:11:42,209.72948838 --> 00:11:46,409.72948838
Racking up eye watering bills or flooding systems with absolute nonsense.
111
00:11:47,99.72948838 --> 00:11:55,649.72948838
These digital helpers can hallucinate facts, make embarrassingly basic errors and suddenly hit brick walls when they reach their context limits.
112
00:11:56,339.72948838 --> 00:11:58,949.72948838
And let's not forget the legal and privacy landmines.
113
00:11:59,369.72948838 --> 00:12:05,69.72948838
Agents might mishandle sensitive data accidentally plagiarize content or produce results.
114
00:12:05,69.72948838 --> 00:12:09,119.72948838
So biased they'd make your HR department break out in a cold sweat.
115
00:12:09,509.72948838 --> 00:12:16,319.72948838
So how can you make these digital helpers truly work for you? Let me break it down with four essential principles.
116
00:12:16,709.72948838 --> 00:12:19,19.72948838
First, give crystal clear instructions.
117
00:12:19,409.72948838 --> 00:12:20,969.72948838
Your agent isn't a mind reader.
118
00:12:21,209.72948838 --> 00:12:23,819.72948838
It needs to know precisely what success looks like.
119
00:12:24,299.72948838 --> 00:12:27,389.72948838
The more specific your direction, the better your results will be.
120
00:12:27,809.72948838 --> 00:12:29,969.72948838
Second, grant the right tool access.
121
00:12:30,389.72948838 --> 00:12:38,909.72948838
Think carefully about which systems your agent needs to interact with, whether that's email, calendars, or databases, and set proper permissions.
122
00:12:39,539.72948838 --> 00:12:44,879.72948838
Third, prepare for hiccups with solid error handling if or when things will go sideways.
123
00:12:45,569.72948838 --> 00:12:49,799.72948838
And finally, perhaps most importantly, maintain human oversight.
124
00:12:50,279.72948838 --> 00:12:54,779.72948838
Always create checkpoints where actual humans review the work before it goes live.
125
00:12:56,450.0690649 --> 00:12:57,170.0690649
Let's wrap up.
126
00:12:57,830.0690649 --> 00:12:59,750.0690649
AI agents aren't science fiction.
127
00:12:59,840.0690649 --> 00:13:06,110.0690649
They're today as digital interns tirelessly working away on real tasks with varying degrees of imagination and flair.
128
00:13:06,530.0690649 --> 00:13:20,600.0690649
They're not magical solutions, nor are they job stealing monsters, and they certainly haven't yet replaced good old human experience or common sense, but they are revolutionizing how we delegate, automate, and achieve results.
129
00:13:20,990.0690649 --> 00:13:25,640.0690649
If you're curious enough to explore their potential, cautious enough to verify their work.
130
00:13:25,955.0690649 --> 00:13:28,655.0690649
And bold enough to experiment with what they can do.
131
00:13:29,555.0690649 --> 00:13:30,815.0690649
And here's the brilliant part.
132
00:13:31,145.0690649 --> 00:13:33,275.0690649
You don't need a PhD to get started.
133
00:13:33,635.0690649 --> 00:13:35,495.0690649
Simply begin with small tasks.
134
00:13:35,675.0690649 --> 00:13:42,515.0690649
Utilize those convenient plug and play tools and keep a watchful eye on what they produce with the right expectations.
135
00:13:42,905.0690649 --> 00:13:45,785.0690649
This technological wave can save you countless hours.
136
00:13:46,175.0690649 --> 00:13:51,155.0690649
Add impressive capabilities to your toolkit and free you up for the work that truly matters.
137
00:13:51,665.0690649 --> 00:13:56,270.0690649
So why not give it a go? I'm genuinely excited to see where this journey might take you.
138
00:13:56,630.0690649 --> 00:14:00,560.0690649
If you enjoyed this breakdown, I'd love to hear your stories or questions.
139
00:14:00,950.0690649 --> 00:14:07,490.0690649
What would you like to delegate to your own digital intern? Drop me an email at hello@theaibreakdown.com
140
00:14:07,940.0690649 --> 00:14:11,540.0690649
and follow rate and share if you know someone battling work tedium.
141
00:14:11,960.0690649 --> 00:14:14,750.0690649
Thanks for listening, and I'll catch you next time.