A Look Ahead: Trends in the AI Workforce
While impossible to predict what's coming, it's clear that AI Workers aren't going anywhere. Hear what VC leaders are seeing in the field and where they see the AI Workforce going.
Hi everyone, I'm very excited for this next session.
With me, I have two luminaries in the industry,
Brett Queener and Scott Beechek.
Brett Queener is a managing director at Bonfire Ventures.
He started software career at Seabull Systems,
helping build a juggernaut of alliances and marketing organizations.
In the on-prem world, he then de-camped a salesforce
where he did an amazing job at failing upwards,
his words not mine, for 12 years,
running rev-ops, then product,
then leading several of their larger acquired business units.
Brett then tested his pre-revenue entrepreneurial metal
and lived the hard things about hard things as COO of smart recruiters,
which has subsequently become a unicorn.
And Scott Beechek is a partner at Northwest Adventure Partners,
where he focuses on early to late stage investment opportunities
in enterprise SaaS,
with a particular interest in companies building next-gen systems
of intelligence and automation across e-commerce, sales, pipeline acceleration,
customer service, sales enablement, and CRM.
And so with that, what I'd like to do is kick this over
to Brett with my first question,
which is the concept of AI workers.
It feels really new.
And Brett, I was wondering if you could tell us how you define AI workers
and what they are.
So I have done about 45 interviews
over the last three months about to publish,
hopefully what are some seminal pieces.
But so much is changing.
And so within that construct, the way I think about workers,
I have two camps of workers,
one what I call a digital assistant.
And the way I think about a digital assistant is
think about it replacing the software you use today.
Right?
Software that you use today has a bunch of tabs
that you hunt and peck across and in your head.
You have to translate what you want to do
to learning how the application works and making that happen.
And what's really exciting about using language with digital assistants,
our employees can be so much more effective
when they could just communicate what it is they want to do,
what they want to know in their own works.
And so that's what I think of digital assistants.
And then digital workers in my mind are fully autonomous workers.
And the way to think about it for those that are old,
it's the old ways to think about automation.
Stuff that humans had to do, but got automated away.
And so these would be people that effectively, you know,
digital worker would be, you know,
something that just, there's a piece of work that you do today
that a human actually does that you no longer have to do.
And when I talked to founders where, you know,
used to be like, let's think about your org chart
over the next two years.
When you're going to hire these SDRs, AEs, et cetera,
the question I were now having is,
what is your org chart going to look like in two years?
And it's going to be a mix of humans using assistance,
or we used to call software,
and they're going to be digital workers with humans training them.
And the really cool thing about digital workers is
they have no ego and they'll tell you where they suck
and they'll say, train me, I'm not very good at this.
And so that's where I think the orgs are going
and I think it's going much faster than people recognize.
- I think that's a really important topic,
the speed at which change is happening right now.
And Scott, I would kick this same question over to you.
How do you think about AI workers and what they are?
Like, you see a lot of companies in the Northwest portfolio
and all the companies pitching to you.
What are your thoughts here?
- You know, nowadays, Robert,
I think every company thinks of themselves
as an AI company in some way,
and AI is nothing new, we know that.
I mean, we've had neural networks mathematically
since the '60s and '70s.
But with this latest generation of technology
that really reared its head publicly in November of 2022,
now this has unlocked a lot of capabilities
to build the type of digital assistance
that Brett was alluding to earlier.
I think what's happening is that we're moving faster
than a lot of us thought.
And I think there's a lot of companies,
in the Fortune 1000 and beyond,
who drastically changed their course
or were being told by their CEOs and other leadership
to embrace this new technology.
And the question, the early questions were,
well, what the heck can we do with it all?
If chat GPT was an example of this,
could we use chat GPT to automate some of the simple tasks
that we were doing in our jobs,
like writing emails or writing code?
Or one-dimensional type of tasks,
doing a little bit of research here or there?
I think the answer was pretty clear, it was yes.
But now we're starting to get into this phase in 2024,
where we're not just using single threaded
type of digital assistance.
Now we're stringing together digital assistance
and creating digital AI agents.
Agents that are now capable of not just doing
what you might consider tier one type of work,
which is the sort of one-shot, one-and-done type of work
to answer a question.
But now we're starting to think of second order type of work.
The type of work that you might see
a customer service agent doing or the type of work
that an SDR might be doing,
where there's a little bit of research involved,
there's maybe some insights that they need to derive
and then some action that they need to take.
And I think we are now in the AI agent world
and we're quickly moving beyond that
into something that we can now, I think, call AI workers,
which is going to be, I think, a big unlock
for this next generation of technology
and productivity for companies.
- Scott, you sort of took my next question on already,
which you talked a little bit about the timeline
and then what's happening with the speed of transition today.
Brett, from your perspective and the companies
that you were seeing,
based on what you saw in the beginning
to the products that you're investing now,
what's that transition been like?
- You know, I was just talking to a guy we all know.
Will Moxley used to work at Seval,
was Ransales Cloud at Salesforce,
now runs product at Apfolio,
$600 million property management software.
And instead of replacing what I call a simple crud action,
create, what does it create, read, update, delete.
If somebody asks something like, hey,
is there a storm coming in a certain area
and the answer would be yes,
instead of having to go through the application
and query which properties, put some bully in,
then figure out like, okay, who are the points of contact?
What language they speak in?
The person can now say, hey, for the hurricane
that's coming to Tennessee,
can you please just notify everybody there,
some message of like, get out or bat in the hatches, done.
Like from a user perspective, that fundamentally changes
the entire way somebody thinks about using that software.
And then, what does the agent ask?
Would you like me to do that for you every single time?
We have unchanged somebody who has a job
to be a property manager from hunting
and packing across an application
for this computer or agent, if you will,
doing this on their behalf,
and when you see that in the orgs, they're stunned, right?
Because what is happening is that the software,
whether it's an agent or assistant,
is actually really helpful to the individual.
They don't need to be trained,
I don't need to do six months of onboarding, right?
If the product works in the language
to solve whatever problems somebody wants done,
what is onboarding?
So, it's pretty exciting,
and I would say what's happening now
is orgs start to understand the power of these LLMs
if they have the right architecture,
and they've trained these LLMs on their metadata.
The ability for these agents,
either as an assistant agent to do,
what we would think of a multi-chain set of CRUD actions,
it just does, is mind blowing.
And so, I think for the R orgs that start to recognize
that we have an interesting word that we use,
which is minor miracles.
R and D teams are finding minor miracles
like every three weeks, they're like, "Oh, holy shit,
"we can do this now."
And then when they show that to customers,
it just fundamentally changes the value you have
as a technology provider,
whether it's a worker, as an agent, to these organizations.
And it also brings up a really interesting
budgeting conversation, right?
On the assistant, it's gonna be measured on,
do I make the function worker that much more productive?
And we know that we do.
On the worker, it becomes interesting,
because that's really about replacing payroll spend,
or redirecting payroll spend you had
to hire value activities.
And so, these are all really new interesting concepts
for the three of us who've been enterprise software
for way too long, that we actually,
all the playbooks that we have before,
we kind of have to chuck out.
We'd be like, "Well,
"so anybody has long answer to a short question,
"but I have never been more excited
"in terms of how we've all been building software
"to tell companies we're gonna make them more productive
"and more successful and the end users will love it."
And it's actually really now possible,
and it's super exciting.
- Let's pull on that thread a little bit, Scott,
because I think this is getting into an area
that's really interesting around end users.
But why do you think AI workers offer so much potential?
What are they really good at?
And where do you think they're gonna help us the most?
- Well, I think one of the first things to keep in mind
is that AI workers, we are building and training them
to act and behave a lot like human workers.
But it's pretty clear that we're not gonna replace
all human workers anytime soon.
So there needs to be a world where we coexist,
and we work with them.
And that's like a form of collaboration.
What's interesting about this is that AI workers
are gonna work with humans,
but they're also gonna work with other AI workers.
And that creates another interesting scenario
called orchestration, because AI workers,
they speak their own language internally.
And so now there's a whole new category of companies
that are allowing agents and AI workers
to work collaboratively among themselves
and then also work collaboratively with humans.
And so I think the answer to your question is,
what we're seeing right now is the first generation
of AI workers are taking on those tier one tasks.
They're taking on tasks like outbound sales outreach.
An inbound sales handling of customer leads
and customer interest, they're handling tier one customer
support where there may be the top 10 or top 15 questions
that get asked repeatedly of a call center agent
that can handle those pretty well.
And by the way, the amount of value
in just solving those three use cases,
and by the way, there's dozens more.
You look at the back office and automating tasks
like simple accounting tasks,
or you look at the idea of closing the books
and how you can automate some of those things
with some automated, you know, CPA like workers.
What is going to change the world, I think,
is when we start to integrate these workers
among the rest of the employees in a company
and watching them interact and become,
making not only the processes more efficient,
but actually making us as humans more efficient.
Because if we can offload tier one work
to these other workers, imagine what that unlocks for us.
As hopefully, tier two, tier three, tier four,
I'm mostly speaking for the two of you
on these higher order thinking measures,
but you can see, this is a very real,
I think when AI really started to become prolific,
there was a real concern, you read it in the news,
oh my gosh, it's gonna take over all human knowledge work.
Well, that's not what's happening,
and that's not gonna happen.
What is gonna happen though,
is it's going to allow us to do higher order work,
which tends to be much more interesting.
And are there going to be human beings on this planet
who get automated by AI workers?
Yes, there will be, but guess what?
Those people now have an opportunity to learn and grow
and have a more enriched and more fulfilling
work experience alongside AI workers.
Scott, I really like the way that you're thinking about this.
And, Brett, I'll pose this question to you.
As we start looking at this from a broader perspective,
what trends are you seeing in this market?
And are there particular areas of a business
that you've seen where the most influx of AI workers
being created or an area that you're particularly excited about?
I saw a demo the other day that kind of blew me away.
You asked about what this stuff is good at.
It looked, don't call it an agent,
but if you look at GAL, if we think about like three years ago,
we didn't do call recordings, right?
And now the idea of Scott, you and I meeting with the founder
and taking notes and not having the call recording
and not being present on the call
and then having to update our partners
and remembering what it was
and forgetting to call the founder back
and then you're seen as a bad PC
and then trying to update your,
like what the hell was that?
That's what we all did.
You would never not do that now.
And so what's very interesting is,
what is AI really good at?
That humans aren't good at?
Is looking at large set of data
and providing some synthesis.
And GAL was sort of our first history of that.
But like, if you're a product marketer today,
that job has changed dramatically.
'Cause now I can listen to every call,
every conversation, look at all the data
and I can figure out what's resonating,
what's not resonated.
We don't spend six months figuring what an ICP is, right?
Like if I'm a product manager now,
like there's enough data out there
that an agent can tell me based on all the stuff
what's going on.
So, but like a really cool demo I saw the other day
in these agent building and orchestration platforms was,
you can easily build an agent.
I'm a worker and say,
"I want you to be the customer success worker
"and you work for me, I'm the CEO."
And I want you to just go look at all the customer success stuff
and tell me what you think the 10 biggest product issues
that is coming that we have from our customers.
Just tell me.
And then you can create one which is a product agent.
Hey, you're the product agent.
As I want you to look across Gura tickets
and all I can tell me like,
what are we actually releasing
and what do we think is coming out, et cetera.
That's kind of interesting.
But the one is really interesting,
which is I create an agent,
so you're the meeting agent
and I want you to meet the product worker
and I want you to meet the customer success worker.
And I train the meeting as I want you to figure out for me
where these two words are most disconnected.
And you know what?
That data is all there and it could tell you,
oh my God, if you're running a function
or you're running a CEO and you spend all your time
just trying to figure out where the disconnects
from your orgs are.
And then if you get promoted or large
like Scott and Robert were where they were like
EVPs and presidents and sometimes they had to go
like put a disguise on a go meet with the ICs
because once you talk to the VP's and the EVPs
and the EVPs you actually don't know what's going on.
That's amazing.
So then you as a leader can actually have that information.
It's been synthesized and then you do what you do well,
which is problem solve.
So like when you do that,
it kind of just blows you away
and it's not hard to go do.
I think we're living in a world where data
has become so important and that synthesis of that data
and what you just described, Brett,
is something that's really exciting.
Scott, is there an area that you're most impressed with
and is there an area that you think AI workers
still need improvement?
Yeah, well, Brett hit on something pretty important there
which is that metadata is of course a moat,
data becomes an input and a signal to all these systems.
What's happening right now is that we're seeing a proliferation
of the large language models, open AI and anthropic
and cohere and a Mistral and a whole bunch of others
and they're providing a platform on which you can build
your own agents and you can point them at vector databases
and implement RAG and you can fine tune them
in all kinds of different ways
to solve different very specific problems.
But that's not the right answer for a lot of
what will be the future use cases
that companies really want to go after.
In fact, they're very bulky.
I mean, think about it.
Like if you wanted to solve a simple task,
like if you wanted to generate an outbound email
to a sales prospect and you just had a little bit
of information on the prospect you're sending the email to
and the information about the customer,
you probably don't need a two trillion parameter
large language model to do that, right?
And so it's a bit like, you know, overkill.
And so what's happening now is that we're this,
the advent of small language models
and we can build these systems of extraordinary intelligence
and automation.
We can build AI workers that run on the edge
that are gonna run on your mobile device,
that are gonna run locally on your laptop,
that are gonna run very fast and lightweight
on our servers and very inexpensively.
And so what's gonna happen is most companies over time,
think about the next couple, two, three, four years from now,
are going to have hundreds, if not thousands, of AI workers.
Some of those workers are going to be working with people
like us alongside us.
They're going to be working with interacting
with our customers, our customer service and sales.
They're gonna be interacting with our tax firm
to exchange account information or invoices
and things and analyze contracts and so forth.
And those AI workers are sometimes going to be very good
at that thing that they do, but very bad at other things.
We've built so far a lot of AI workers
that are pretty good at pretty much everything.
These built on these massive, massive horizontal
large language models that are not necessary.
And quite frankly, it's not gonna scale over time.
So the answer, I think, to that question is that over time,
AI workers get better and better and they get deeper
and more skilled at that thing that they were built to do,
but they're gonna get worse at all the other things
that they weren't designed to do.
And so I'll just give, you know,
an example here that I think will probably hit home.
If we've got an AI-SDR to get asked to compose
a piece of music similar to a Mozart symphony,
that thing should say, "I'm sorry,
"that's just not what I do."
But I sure as heck am happy to tell you all the reasons
why you should buy our product
and let me give you a demo, right?
That's kind of where we're heading, but we're not there yet.
We're very quickly moving in that direction though.
- So I think in listening to the two of you,
it's spurred so many interesting ideas.
One of the things that I get asked a lot is advice
for companies in terms of what they need to be prioritizing
to be ready for this future.
In conclusion, and I'll stick with you Scott to begin with,
what advice would you give for companies
who are looking to get into this AI workforce space
and what should they be prioritizing right now?
- Yeah, what I would say is don't try to boil the ocean.
Right now, we're at this point,
I think in the technology arc for AI,
where focus and narrowing of scope is really gonna pay off.
So as a company, as you start to think about
how what you really wanna achieve as a business,
work backwards from those goals,
do you want to increase your sales productivity
from 60% to 75%?
Do you want to decrease your average handle time
in your call center from five minutes to four minutes?
If those are the types of things that you wanna do,
then you look to implement verticalized solutions
that can help you do that.
And just as an example, if I'm a construction company
and I've got a bunch of highly skilled agents out there
and I'm an international firm,
I may be taking hundreds of minutes,
if not thousands of minutes of calls per week,
trying to solve some problems.
And there are going to be generic solutions
for all of these problems.
There's gonna be a generic call center agent,
there's gonna be a generic sales agent.
But like you well know, Robert,
the sort of the effort in building the best of these agents
is all about narrowing its focus.
So, you know, just as a plug,
I mean, Piper, for qualified,
one of the first things that it does is it says,
"You know, what are we trying to achieve here?
"Let's narrow down exactly what the topics are
"that are important.
"Let's narrow down what our goals are,
"where do we want, what do we want to achieve together?"
And then the AI worker then completely changes its behavior
and focuses.
And that's why I think if you're a business
that's really looking to use AI and AI workers
to automate more, don't try to boil the ocean
and get AI workers that are trying to solve too much.
Just try to find those solutions that go real deep,
get very smart and just as much metadata and data
as they possibly can to train themselves
and just start small.
Because those small wins will afford you,
you will earn you the ability to go tackle
the next problem and over time,
we're gonna have armies of these AI workers
that just completely transform our entire business
and make us more productive.
- Brett, what are your thoughts around advice
to companies as they look to prioritize?
- Yeah, when I heard the word thousands of AI workers,
I get a little worried, but it's fine.
But what I get worried about is
last 10 years people bought too much software.
The average employee in a functional role
leaves us like five to 12 pieces of software.
And I wanna make sure that we don't repeat that.
You had mentioned before Scott,
the importance for orchestration, right?
And I just see people going crazy with a bunch of workers
not being thoughtful, the rest of it rolling it out.
And it's just now I've got a bunch of software
that doesn't talk to each other,
a bunch of agents don't talk to each other,
a bunch of employees don't talk together.
And then people are like, "Ah, AI doesn't work.
I can see that happening."
My guidance, most of my guidance is with smaller startups.
And what I tell them within the first 10 employees
is don't almost ignore all the playbooks
of how to go build and run a SaaS company,
'cause they're all sort of predicated in a pre-AI world.
And you will go do stuff and you will buy software
the way everyone else has done
and actually have them sit down
and do first principles.
And so I say within the first 10 people,
10 people they hire, I want one to two people.
And they tend not to be people that write Python.
You can hire, there's a reason why Blackstone
is hiring a bunch of liberal arts majors.
You're looking for people that are problem solvers,
first principle thinkers.
And I would have them as part of the leadership team
out of the gate thinking through
as we're building each org and each process.
What are the core processes we're trying to do here?
And where do we think AI will provide the most leverage?
And as Scott said, let's start with a couple
where it's relatively focused where we think
that this agent can do a better job
than a human for this, not a cheaper job.
You know, one of the challenges we use
where agents a lot, people always couldn't
and respond weird to the word agent
because they're used to bots
that run customer service websites
that were really bad, right?
There was no real intelligence,
it was just the FAQ for call deflection.
And so it's very interesting, we have to overcome that.
And so I always say, I want you to do this better
than a human and I'll give you an example.
We write a lot of content for our founders.
We hold podcasts, we get speakers.
I think they never show up and they never read it.
So we're sitting on a call like,
oh, maybe we should just stop doing this.
Why do we do this?
And I was like, well, because if they read it
or they did it, we would have scale.
So I was upset and I was like, forget it,
I'm gonna create an agent.
I'm gonna create a digital worker.
It was like Wednesday at three o'clock.
And no, I'd do it.
But like at nine o'clock the next day,
I had adjusted all of our content,
found something to rip the podcasts.
I think you used to chat base, it's a hundred bucks.
And I'd create As Bonfire in 30 minutes.
Now I had to train it, like, you know,
it came back and it had an answer,
but it didn't have the link to the doc.
And I was like, hey, dummy,
when I've given you a document
where there's a link, please provide it, train.
And now that's what I give to the founders.
That took me like 30 minutes.
But what I did, which was interesting,
and when I first did it, I grabbed all of our content.
But a lot of our content actually was brought,
like we invested in this company
and this goal was to be successful for founders.
So that we took all the content out
that wasn't around advice for founders
and it's pretty damn good.
And we did that like 30 minutes.
I didn't write any code.
And then like, okay, how do I expose it to people?
Well, we put it in our Slack group.
Now, they don't ever come to the Slack group.
That's a whole separate issue for another day.
But that's the opportunity of thinking,
how to think about it.
But I would stay focused.
'Cause I also think what Scott wrote about is,
this feels uncomfortable at first.
But once you start playing around with it
and you start doing it, you realize,
oh wow, and actually your approach to AI
and how you think about it changes.
And so, but you have to build those wins
within your company and the belief.
So start with like the high value use cases
that are very focused and then it'll become,
just like you said, we don't think about AI
as something separate, all software companies are AI.
That's kind of the guidance and the approach
that I give people.
I like to give you an example.
And Scott, you'll love this from your Salesforce Service
Cloud days.
People are like, okay, I've got this
agentic interface in my product.
Okay, great, wonderful.
And it's helping you use the product, et cetera.
I said, if somebody wants to file a bug
or log a ticket or has a question, where do they go?
Oh, they go to this other tab over here
to go interface with that.
And I was like, I don't understand.
You were just having a conversation
in your product with the user.
Why wouldn't it just be in there
and the user would just ask that?
And people go, oh yeah, why wouldn't it be that way?
But you have to sort of, we have to untrain ourselves
in terms of all the shitty ways we used to do stuff
and be like, that doesn't make sense.
So that's kind of the guidance I give folks.
- Well, Brett, Scott, I want to thank you both
for your time and your insights.
This has been a great event.
We're excited to actually host the next set of speakers
that are coming on.
Thanks again to both of you.
Really appreciate the support and Q out.