SaaS Backwards - Reverse Engineering SaaS Success

Ep. 163 - Is Your SaaS Ready for the Agentic AI Era?

Ken Lempit Season 4 Episode 16

Guests: Ken Lempit, James Ollerenshaw, and Rob Curtis

AI isn’t just a bolt-on anymore—it’s rewriting the rules of SaaS from the ground up.

In this episode, Standup Hiro co-founder Rob Curtis and tech strategist James Ollerenshaw join host and GTM expert Ken Lempit to unpack how Agentic AI is forcing SaaS leaders to rethink everything: product design, go-to-market, customer trust, and even internal culture.

From natural language interfaces to the rise of AI “co-workers,” they break down why SaaS companies that adapt will dominate—and why bolt-on AI features won't be enough to survive.

📌 Key Takeaways:

  • The Age of AI Co-Workers
    Forget copilots. The next SaaS revolution will be about true AI teammates—systems that work with you, not just for you.
  • Natural Language is the New UI
    SaaS buyers are expecting to talk, not click. Voice-first experiences and conversational commands will soon be table stakes.
  • Trust Is the New Battlefield
    Enterprises are more forgiving of human errors than machine mistakes. To win with AI, SaaS leaders must prove not just capability, but reliability.
  • Bolt-Ons vs. Reinvention
    Tacking AI onto legacy products won’t cut it. Companies willing to re-architect around AI-native principles will seize the biggest opportunities.
  • Invisible Software is Coming
    Many SaaS tools will fade into the background, powering workflows behind the scenes. The battle for customer love (and margin) starts now.

Heads up: This episode runs longer than usual—but for good reason.
We go deep into the real-world implications of Agentic AI, from product strategy to pricing models, trust, and the future of SaaS UX. If you're a SaaS leader trying to stay ahead of the AI curve, every minute is packed with actionable insights that could reshape how you build, sell, and scale.

---

Not Getting Enough Demos?

Your messaging could be turning buyers away before you even get a chance to pitch.

🔗 Get a Free Messaging & Conversion Review

We’ll analyze your website and content through the eyes of your buyers to uncover what’s stopping them from booking a demo. Then, we’ll give you a personalized report with practical recommendations to help you turn more visitors into sales conversations.

And the best part?

💡 It’s completely free.

No commitments, no pressure—just actionable advice to help you book more demos.

Your next demo is just a click away—claim your free review now.

Ken Lempit: Welcome to SaaS Backwards, the podcast that looks at what's working and what isn't in the world of B2B SaaS.

This is the second episode in our new series on the impact of artificial intelligence on the SaaS industry, where we're unpacking what AI really means for builders, sellers, and buyers of enterprise and departmental software solutions.

In our last conversation, we explored how agentic AI and no-code tools are rewriting the SaaS playbook, lowering the barrier to building software, and calling into question how much value SaaS vendors really own.

Today we're diving a little deeper into the world of agents, what they mean for SaaS and what opportunities they present.

Like voice, which we'll cover on next week's episode.

Today we'll be asking questions like, how close are we to a world where we don't have to click through dashboards, but use natural language to drive SaaS apps.

Something like Captain Picard talking to the enterprise.

And what happens when agents are capable of initiating workflows, handling escalations, or even rewriting business logic on the fly.

To explore these questions. I'm joined by two brilliant minds, James Ollerenshaw, a marketing strategist with deep AI experience and healthy skepticism earned from watching more than one hype cycle play out. 

And Rob Curtis, co-founder of Hiro studio, A venture builder working at the cutting edge of Agentic AI and a frequent chronicler of this moment in tech via his substack.

I am Ken Lempit, I'm your host, but today I'm gonna turn hosting responsibility over to Rob Curtis, and Rob, take us into the episode.

Rob Curtis: James, Ken, great

to be here 

talking about what feels like the future, but is

candidly happening right now.

This week we're talking

About agents.

Agents have been in

the news this week quite a lot,

and I wanted to share

two stories that give very different signals

about agents, but I think perhaps present both cautionary tales and some optimism for this audience.

I don't know if you saw the ads 

stop hiring people.

Large billboards, 

Times Square, SF. They went viral.

They spoke to

the deepest fear of being 

replaced by our creations. That company

Artisan has just raised its next funding round.

There's a lot of conversation going on

at the moment about how we 

can take out more human cost and replace it

with more automation.

At the same time, 11x maker of more synthetic human workers,

especially SDRs, has been sued by

ZoomInfo for using their logo at the end of what I

believe was a failed pilot.

They're also being looked into after allegations have emerged from their staff, that they've been inflating their retention numbers. 

The company is saying something seventies to

eighties percent retention. And their staff are saying these are significantly below that.

Why does this matter? Not only are we all in the market

for SDRs but it tells a story about what investors

are investing in and what the

world is looking for, which is

more effective ways to use the tools

to achieve the same outcomes. And that's going

to affect software creators like listeners of this podcast.

Like you said, Ken

starts to feel like

the Star Trek Enterprise is upon us.

I love the idea of using natural language. We trained ourselves to talk like robots to Google, Restaurant near me, Vietnamese vegan.

Nobody talks like that.

 With the rise of

LLMs and chat interfaces, we are starting to talk again like humans.

Ken, you've been around a while through the world of SaaS. We are

now in a world where natural language allows us to operate our SaaS products,

potentially initiate new systems,

log in, make payments, do a whole range of

things in natural language.

One example is the difference between looking in retool for your data

Versus asking a simple question, analyze Q4 performance. 

Ken, as you look around your clients and this industry, 

how ready do you think SaaS providers are for the world of

natural language?

Ken Lempit: So I think a lot of this depends on where they are in their development and

who they sell to. We're working with a client in 

the manufacturing software

space and their product

was built decades ago.

The original code is probably 20

years old and it's being reinvented on an AI

native platform. The most exciting part isn't the dialogue

like interface. It's a pretty

plain Jane interface to operate

the mainline function of the system. It is actually the ability to inquire of this very

complex data set in natural language.

Companies 

prepared to reinvent themselves or are

in launch at the moment are the ones

that are gonna be most likely to

lead with these natural language 

capabilities.

And it's really exciting. 

To see

data able to be accessed

in ways

that were unimaginable only two

years ago. like you could not imagine doing what

you can do today

in a command

line or chat 

interface.

There's

gonna be opportunity where folks

have great data.

Like their data model's solid and they can reveal

their data to an AI interface. And I think

 The posture's gonna be most 

aggressive with young companies either, 

reinventing themselves

or, 

starting fresh. 

Though

If you have a great data lake and

you're willing

to expose it to an agentic

or natural language

interface maybe you can have stuff that's just as exciting as what we saw only a couple of weeks ago in this startup client's demo.

Rob Curtis: Thanks, Ken. That's super interesting. And James, you are one person that often

reminds us of the expectations of enterprise buyers,

and in particular around things like accuracy.

In a world

of natural language versus

a world of

say, command prompts,

It's 

gonna

be really interesting to see which of those is most capable of bringing out the most accurate instructions to lead to the best results.

What are your thoughts? 

James Ollerenshaw: Before I go into that question, 

I want to go back to

what we were

saying about us entering 

the

world of Star Trek

And the enterprise talking to the computer in,

Natural language I think that's an

incredible

thing to have happened.

That shows 30 years old or something and

It was real, science fiction

to say computer and have this conversation.

The computer had to converse back

and help us solve problems. We work with the computer in real time using natural voice. Ken, you set me up as the skeptic here.

I have been through 

a couple of hype cycles

with AI

and see where this can be a challenge

 But I'm a

great enthusiast for it. I've got my Star Trek necklace on so I'm all in on this kind of future working.

But it's gotta work. Does it work effectively? If you are, 

In 

an enterprise and

you have a process,

that process

needs to be

reliable. What we see 

with the latest

wave of

AI is that it is not always reliable. It hallucinates it interprets instructions different ways depending

on how they're expressed.

We see that in typing into ChatGPT

interfaces, now we're able to

talk to the computer.

That creates 

More scope

for variation in how

we create those

commands. 

And even

more scope for variation in how the computer responds.

That can be wonderful 

but

when trying to get consistent outcomes,

it's not always helpful.

As

we start

to talk to computers more,

how can we be

confident we

are getting

consistent results?

Not always the

same results, but consistent quality, integrity of data

fairness of customer

data protection. These are 

very real 

questions for those implementing these tools 

into processes.

Everybody's excited about buying agents 

and not hiring people. I still see a great need for the human in the loop 

Rob Curtis: I wanna bring it back to natural language. The question we kicked off with was, how do we feel natural language versus prompts is gonna bring out the best in these systems? Do you have a take on, where the line between

a prompt that's spoken like a human versus Q4 results variances now?

What do you think is going to drive the best results? 

James Ollerenshaw: That's a good question. It's a difficult one

to answer. It depends. 

 I've 

been involved in AI with regulated

industries. Particularly financial services,

to some extent, healthcare as well. These tools are not reliable enough there to be

used without

significant human intervention. When interacting with Gemini or ChatGPT, we put in the prompt, see the result.

And if we are educated enough and

paying attention enough to the

output, 

we can determine it's quality.

But as that gets much faster 

it's new voice and we have less time, maybe the people we put to work on that are less educated in a particular process and outcomes and what good looks like that question of reliability and risk of errors becomes greater.

So, the need

to have

capable humans in the

loop becomes greater. It depends

on the industry and tasks. If we're trying to sell holidays,

And having

A conversation with a computer about,

Where

might be nice to go, can the

agent then go and actually book things,

for me, that's a much lower risk situation than managing my bank account. Or the results of a Cancer scan. Companies

need to think about this carefully.

 In many cases regulations will prevent application of these. But

even if it's booking a holiday, you don't want it to make a mistake. You want the car available on the correct date, 

the hotel to the preferences you've set. 

So, I think the capabilities are exciting, but

It might move

a little slower

in reality

than many of the vendors would have us think. 

Rob Curtis: Yeah.

As 

I think about the stories 

we read about the

technology,

one that comes to

mind 

is using pleases and thank yous with

LLMs. The insertion of a single word can change 

the attitude of the LLM and its outputs.

These are probabilistic models, so one could imagine

that any particular string

of words, will create a perhaps different presentational answer.

But whether it actually

presents a

different

conclusion is to be

seen. This is

going to

force us to think carefully about how we use

natural language or force the platform

players to

be able to take stream of consciousness and turn that into reliable

prompts that drive

the software in its intended fashion.

 I want to go back to Ken, something that

Came up in our last

episode.

You predicted

last

week that many SaaS leaders are going

to have

to rethink their platforms for the agent age.

I'm seeing some real reinvention and some

Bolt-on

AI assistance.

 Ken, what's your

take on how this is playing out so far?

Ken Lempit: I think the bolt-ons are

a great way to get 

Experience, what I like

to call

time over target.

They don't require

a complete rethinking of a product to start

getting some

traction. If you've got

a hundred million

or more in revenue on a SaaS

platform, it's not like you're just gonna snap your fingers and change direction right away. 

You need some 

market confirmation

that the things you'd like to build are gonna be valuable. 

I think as a user organization and user individual the co-pilots

are probably less valuable than

to the vendors

themselves,

right?

This is the way they figure out how people want to engage with their product in new ways. But they don't seem tempting to me. 

Not in my

experience. 

Depends on what that copilot is doing. 

If you're a user in an area where you have to do repetitive data entry, I guess a copilot could be

very attractive. Copilots are, 

for the most part, an interim part

of this migration to smarter software. There are probably gonna be

applications

where they become more robust and do 60,

70, 80% of the work and then ask for

permission to complete a project or process.

I think that's what you're gonna see next 

is this, let me help you 

with this

document

categorization project, and

I'm gonna

confirm along the way that I've done it right. There are also gonna

be 

applications where 

the

risk isn't as

great and

we're gonna let

these processes

rip and do a little QC

and make sure they weren't stupid.

I don't

know if there is a definitive answer here. It's almost like human experience, almost anything we can imagine as a

way of working

is gonna be represented by

co-pilots and semi-autonomous agents.

Rob Curtis: Ken, as you say that, it makes me think about the use cases

in which people are using agents in their businesses right now. For example, Cursor or Replit or any number of these coding apps, the app is there to drive the software. It is to bridge the gap between a complex system and a user

who might need help, guides

support, learning and education in order to use that product. Using co-pilots in

those instances is a game changer. That's where Vibe Coding

has ultimately come from. And you see these popping

up more like copilots as a way of improving product

usage, depth, and retention.

Then you've got a second set of

agents, which I'll just throw out

operator by OpenAI as a good example, their job is to connect

workflows across multiple systems. What's the 

purpose of that agent?

Is sticking

an agent on your knowledge base enough? Or is your desire to integrate your platform into a wider ecosystem of other platforms?

Is that where your agent

strategy comes in? 'cause those are

Very

different ways

of using a similar technology.

And James, so I am curious to, to hear what

you think. 

James Ollerenshaw: Yeah, I think it's

the data

Behind that agent as well. 

It's easy enough to stick on bolt on. Chat GPT 

or Gemini onto your

existing SaaS platform

and get 

time over target, 

with minimal development change.

But if you are

building, AI native in your

whole data setup 

is engineered for

those kinds of interaction

that AI systems can

have and start to apply in agentic 

ways, that's more towards what

you were saying Rob. It's

a second order of benefits.

That's a significant re-engineering of

the underlying AI technology

that creates those capabilities. So there's value in the bolt on

to explore my data in new ways and get more out of it, but I don't think it's a game changer.

If I have a system where

all the data is set up for AI to

explore and work with it in that way and then go off

to carry out tasks and talk, through APIs to other systems which are

also set up that way, that's when things

start to look very different.

Ken Lempit: I think we need some other language here. The

co-pilot is the bolt-on, I'm using it almost in a 

pejorative sense. There's room to

distinguish between things that are suggesting the next sentence in my email, 

which is of low value

and what you were 

describing as truly a co-worker.

Co-working

with this intelligent agent as opposed to it just trying to augment a little bit my experience.

So maybe we need

Co-worker as 

our thought process, even though

I don't think

we're gonna change the vernacular here on

the podcast, 

but for the purpose of the podcast co-working and being my partner in

execution is where the added value comes.

Right?

Rob Curtis: I think that's where you 

start to see meaningful time and cost savings. 

An SDR

is somebody

Operating multiple platforms, connecting the data on those together and

driving a computer. There are other human skills that come along to it, but some

of the synthetic, 

co-worker platforms are designed to take away a lot of that kind of intersystem interactions.

'cause those are non-strategic work.

I think we've all seen in the media this week, the announcement from the Shopify CEO. 

There was a leaked memo. He recently told

managers individuals will be evaluated on how well they operate AI. And indeed when it comes to things like headcount and new spend, the question will be how have

you exhausted your options on AI

before we start even considering bringing on a new human worker?

A lot of the coverage has been about the replacement of humans. I have a slightly

cynical view on this, which is that 

public companies have two 

audiences to take care of that many private companies don't. One is public market stakeholders.

Look no CEO leaks, anything. SaaS

CEOs now know any internal memo is probably

going to make it to the media. And I would

posit that Shopify is, speaking to these two audiences, 

public market stakeholders, and the next batch of Y Combinator

startup founders to say, we are taking 

AI very seriously.

Here are great signals as to

why we should maintain our

stock price.

But if you are thinking of coming up behind us and

replatforming us as AI native,

then we are also ready to

take you on.

Because those are signals every CEO is

expected to put out to the market.

I've got an AI plan and this is not going

to affect my business. If You're using public comms to

make that point, rather than software tools and

releases, then perhaps that speaks to a gap

between what Shopify is capable of

doing today and what they need to do, which is

get ready to become more AI native

to fend off competitors.

 Ken, when

you think about this from Shopify, 

does it raise questions about their

ability to compete with

AI natives, or do you think this

is just part of the maturity

curve that

all SaaS companies will eventually go

on?

Ken Lempit: That's interesting.

I hadn't thought about it as revealing weakness before. I thought of it

as a heartless point

of view. Not illogical, it was a logical

extension of, I've got a couple thousand employees. I don't want 

10% more. 

How do I coax my teams to align on that mission

of not growing my headcount while growing

Revenue and market share?

So this seemed like a corrosive message. At the same time, it could be argued that as an employee,

I wanna be on the leading edge. I don't want to be a victim.

I want to be a victor

in the AI skills race long as it's not a race to the bottom. So from my standpoint, it was more about culture of the organization, the experience

of employees, and how do you put that kind of message in front of people?

I guess my style would've been to encourage them

to build their skillset

for their own

benefit as well as our company.

And finally, 

our customers, which might be something. James, you may 

have heard from me recently

on the product front, I think Shopify is a dominant commerce platform

Just under Enterprise and down. I don't think Nike's

gonna build its

store on Shopify though.

 I think they're in

a powerful position. And this might be the kind of

posturing that if

there is a weakness in

the product roadmap, would mask that

for a time. 

I don't think that they're

in a really exposed

position, but it's not like I've

done a deep dive on them.

They seem dominant. That's my take. I think it's 

a cultural issue. 

Culture eats strategy for breakfast. That's what we used to say. And I'm gonna say culture eats AI for breakfast. If you have two companies, they do

very similar things.

And one, 

the culture 

is much more supportive and healthy and the other is much more militaristic and, top down and they have the same tech stack inside them. 

The one with great 

culture's gonna win.

I just don't think

You can beat people to success. You have to encourage and grow them to success. 

Rob Curtis: James

Companies are in a bind of managing all of their different stakeholders and they've both

got to satisfy

investors who want to know that

they've got a

hand on things, but also

very rapidly take their

teams through what's gonna

feel like

a very 

aggressive change management process.

And I wonder 

if you've

got any ideas about

how can a leader

balance those things? What

can a leader say to a team about AI that

is not going to have a 

kind of toxic

culture implication? 

James Ollerenshaw: If you take that message at

face value, it was blunt, but 

a call for upskilling with AI. 

The quite famous quote, AI won't take your job, but a

person using AI will

or some version of that which was Richard Baldwin at the 

2023 World Economic

Forum Growth Summit.

We see that happen. If you're able to use

these tools effectively, you are a more valuable

and efficient employee than somebody that can't.

Every

company boss has got to adjust to this.

Rob to what you were saying earlier, the

company bosses have got

to signal to their investors that they are 

engineering their companies and their

employee base that way. I know two perspectives on that. The long view is this is an industrial revolution, machines take on new human skills.

We've just seen another wave of that. 

And the people that can work with

those machines can adapt to that future. 

Probably you're not

gonna take away work because people

figure out other things to do 

it can be messy.

What I've seen happen in practice and companies that I've worked for building, selling

AI and the customers buying that is that they haven't

laid anybody off.

Now I've tended

to be working mostly

FinTech. This doesn't represent all areas of the market, and

things are moving very

fast. The technology comes in and people, 

gain 

efficiencies and better 

outcomes, and use existing teams more effectively. So it may mean less

hiring. It

doesn't necessarily mean layoffs. And I think there's a message

from the Shopify leader in that it's to be able to

stay with

this company, we need you to use this technology effectively. That's a great transferable skill for any leader

or employee.

Rob Curtis: White collar work across many industry verticals is going to be affected. I don't envy the leaders that

have to message that to their

teams, while also messaging to external stakeholders and

The best talent in the market, come to us because we are

AI forward

is a very difficult message to also partner with,

stay with us because we are AI forward, and here is how we're going to support you, help you invest in you. And I would say the Shopify

CEO's note was

a reality.

Any staff member

that was surprised 

probably had not read the tea leaves about their own culture because it's very innovative but I would've loved to have seen, we care about our talent. Here are the things we're doing to protect and

build our talent to be ready for these new

technologies.

I have led teams through

major macroeconomic change, balancing both of those things is incredibly hard and protecting

culture in the face of major disruption and in particular significant shareholder, like change as in what they need. That's tough.

James, I want kinda

go back now

to the impact of agents on buyers in particular. 

Let's think about some of these agents that work across different platforms. I wonder

what you think it will take for enterprises to trust an agent that is as fallible

or more so than a human

in driving complex processes across systems.

How do you think enterprises think about it?

Are they comparing them to humans? Are they comparing them

to perfection? What do you think that dynamic looks like? 

James Ollerenshaw: People do compare AI to humans. Most definitely.

 Humans are intolerant

of the machine getting it

wrong.

It freaks people out when they see a small error rate by

a computer. I've seen this with 

a company that I used

to work in serving the mortgage industry handling application data and turning documents into data and seeing 50 examples of the machine getting it wrong.

And everybody was freaking out about this. And the total number of data points was thousands.

We were talking about an error rate of a small

fraction of a percent

compared to 15, 20% by people, which the industry was entirely

used to and comfortable with.

So the fear factor around

a machine getting it wrong, it's just higher.

So was it gonna take, 

Rob Curtis: why for these, all these smart people focused on math, focused on metrics? Are they wrong?

That is a bad decision. 

Why are they getting it

wrong? 

James Ollerenshaw: Why don't we

have self-driving cars? A lot of the self-driving car technology is already good enough and safer 

than humans driving,

but we don't trust it.

Rob Curtis: We've got Waymo in a bunch of states, so I think we are starting to trust it. 

James Ollerenshaw: People are adjusting, but do

you see widespread

trust in AI? I don't.

Bit by bit, we

start to trust the machines more once they've proven themselves tools. 

 But we have

a very high barrier to this.

If you are in an enterprise and running an operation you've established certain outcomes using human teams.

 It makes no sense. It's interesting

to see the intolerance

or the smaller error rates that machines can deliver against humans. 

Ken Lempit: I think we have a business opportunity here, gentlemen. If we create a standards

Body around the effectiveness of the execution of a Agentic AI, you

could benchmark your application against that standard of performance.

Then you could have

different price levels for

more high quality execution against an agreed standard. Now, that might not be

possible in every scenario, but self-driving is

a great example.

How safe is it

compared to the average taxi or Uber driver? You don't see that kind of

institutional quality assessment that you

do in other areas of commercial endeavor.

So that might be

one way to address it, is to hold

agents accountable,

just like we hold humans accountable.

Somebody's gonna be looking at 

your work output and holding you accountable. James, you had 35 title errors over 10,000 transactions.

That's a great job, but what if you had 300? Would it still be a great job? 

We don't have any external measure of quality of

these things. As a buyer of this technology, how do you discern,

Is Oracle's agentic,

ERP, better

or worse than SAPs, almost impossible to know.

So maybe there's an opportunity

in front of organizations to start

characterizing and calibrating their agents and giving transparency into the results.

If you're tasked with buying an agentic ERP for

manufacturing, how do you know

how good it's gonna be?

I've got,

Fred 

and Mary over here, 

they rarely, if ever, make a mistake. How do I know when I create an automated replacement for 80 or 90% of their work, what is that quality gonna be like? That's one of the

fears and they're born out when some small percentage of errors happen.

I read a 

thing where a Waymo car was stuck going in circles,

It's sensationalized.

How often does an

Uber driver, play with his phone hit the curb and have a flat tire? I bet

it's more often. But it's not sensationalized.

Rob Curtis: If you take it out of life and death examples

and you just think about 

performance is a fiduciary, right?

If confirmation bias led

you to spend more money per unit produced,

Your board would have

a difficult conversation, you've got a cognitive bias encouraging you to choose suboptimal solutions that are costing us money.

If we're in a world of 15% versus 1.5%

error rates, and somebody is choosing a 

10 x worse solution over a cognitive bias,

maybe it's time for us to start having a conversation. 

Self-driving cars are to

some extent a trolley problem.

Do you hit six school kids

or do you hit a priest? That is a philosophical question that we have never been able to answer as humanity, because

it's an impossible question. But when we're talking about loan applications,

yes, these can be

life and death in extreme

situations, but fundamentally they're revenue drivers.

It's gonna be really interesting. 

Last week I talked about quality as

the primary is going to switch

and we're going

to be willing to take 80% of the value or accuracy for 20%

of the cost. I'm curious

to see how outcomes play into the evaluation 

going forward?

James Ollerenshaw: Your

logic is impeccable.

The machine outperforms the human

Many times in many cases.

But my experience of seeing these systems deployed

is that's not

how those buying or deploying them feel about it. I can't mount a logical argument against that.

I've been on the side of the company selling the technology and

arguing the case

that, look, the machine's

demonstrable,

Factually better. I mean Ken you're talking about having ways of measuring it. Side

by side real data

run in tests.

the sensitivity to the machine getting it wrong is way higher.

Maybe humans are kind to humans. 

 It's just gonna take us some

time to adapt.

We completely

trust our dishwasher to do its

job,

but I'm in Mexico at the moment and people

here are suspicious of dishwashers, 

you just don't see them as much. So there is not

as trust, as much

trust in that very familiar machine that we all use every day. it's

A matter of familiarity and 

experience with it. And over

time we can expect

that to change. 

Ken Lempit: So I think the dishwasher's a 

cool analogy and

I'll tell you why.

First of all, I think

I do a better job than the dishwasher. 

I may not be as water or time efficient, 

but my quality is higher. 

But you bring up embodied versus

disembodied, right? The software agent is a disembodied

unknowable. You can't monitor it in

the same way as I can open the dishwasher

after an hour and ten minutes, I can see

very quickly how well it did. When these agents are

embodied, when I can sit side by side with a robot,

It becomes

more like a colleague and less like a mystery.

Now it might

also be more threatening,

right? Which is why a

lot of these embodied humanoid robots are short, to

make them less threatening. 

The embodied

agent, even if it doesn't need to be embodied,

might increase the

acceptance of these co-workers.

They really are co-workers

in a lot of ways. They

may also be more capable of processing,

integrating, and acting on

our behalf. 

The human aspect, 

the organizational change management

is what you were really talking about.

Those things can make or break a project. 

I think it's gonna be organizational change management. Maybe there'll be industry standards for quality of robotic operators

that are needed. 

In financial services transaction processing,

we need to be very accurate, core systems in finance live decades longer than

you'd believe because they're known good.

The

known good process

is of utmost importance.

Could

you imagine if Morgan Stanley said the AI drops the ball 4%

of the time. We'll clean it

up afterwards.

That could be a competitive disadvantage.

It's gonna be that depends 

in terms of adoption.

Rob Curtis: What I'm hearing is that

the way in which

we sell software and we sell

agents is going to have to evolve.

We are going to 

have to be more tenacious. We are going to have to find new ways, not just to sit back

and say, we'll allow the world to catch

up with us. But

I wonder how every salesperson

becomes a therapist to

explain not only that, maybe there is a

bias on the table but how to overcome it. And 

that's our job to sell these products.

And as more come out,

 It's an existential question. We cannot necessarily wait for the world to catch up and we can't afford to ignore

the reality that humans

trust humans in different ways. This is one to watch as AI comes, 

how does it change buyer behavior and what can we

do to influence that further?

I wanna bring us onto our last

topic before we get to predictions,

which is one of my favorite parts of our show each week, there was a LinkedIn or an X

comment from Greg Eisenberg of Late Checkout. Now he does a lot of talking about the future of SaaS and one of the things that he talked about was in relation to agents,

 Two futures for SaaS companies, 

many becoming

agent owners of themselves and operating with the customer,

 Thinking of them as owning the customer relationship and 

owning the customer in

many ways.

And then a second group of software providers that will become invisible software.

They'll be the APIs that agents talk to to get things done. 

A simple consumer example is

OpenTable.

I used to

go to restaurant websites to book a reservation, then I went to OpenTable, and now I never go

to a restaurant landing page ever again.

If I'm going to

Operator

by OpenAI,

which is capable of making

restaurant reservations, it

now goes directly to

the aggregator itself. OpenTable operates all of its features to get me a reservation. If I

ask OpenAI, I've

got a reservation in the table that I want,

at the restaurant that I

want, at the time

that I want. And I think to myself,

will I ever go to

OpenTable again?

For limited use

cases, maybe discoverability, 

but for functional work, probably

never again. I can see this paradigm playing out with many SaaS platforms. If you're asking what are the Q4 reports? Do you need to go into retool? Maybe you don't need

to go into retool as many times. So you start consolidating licenses. 

I'm curious about what agents do to the primary relationship, between a customer and a SaaS provider, and secondarily

is being invisible software a bad thing? James, I'll start with you.

James Ollerenshaw: I think it comes down to reliability.

 Booking the restaurant using

SaaS tools and if that can be a kinda automated through agents

that are using those SaaS tools

and interact with OpenTable, I'll just let it

talk to the computer and let it do that. 

How, much does it matter

to me that I get

a good interaction 

with that restaurant?

So I sometimes find myself wanting to go to the restaurant's, own websites and do the booking myself because

I would've a certain understanding about that restaurant. 

And I also know if I book directly with a restaurant, they're not having to pay a fee

to open table. So I get

better treatment, terms, a better conversation if something goes wrong.

I see that with hotel bookings. 

The systems and incentives that are built into them,

have an effect on all of this. As we trust the machine, what's really going on there?

That's one side of it.

Then Ken, you used 

Morgan Stanley as example, and the

AI getting it wrong 4% of the time, will use people 

to clear up the difference. 

That's what the AI systems of all the different AI FinTech company I've worked with are doing.

We'll put the AI in, it's gonna be more accurate

than the people, but it's gonna fail probably something like

4 or 5% of the time, and we're gonna fill in that gap using people. The challenges is how do we communicate around that to make that seem reasonable.

So I think we'll have

Invisible software that is good enough for most of the time we can stitch up the gaps with people,

but in instances where we care more 

about the relationship that we have, so

the restaurant to the hotel, or the outcome of a process, the visibility 

we need, we might choose to have that visibility.

Rob Curtis: Do you think that's like 90 10? There are things that absolutely cannot fail and 

things that nobody cares about, right? 

The question 

for SaaS providers is gonna be, how many seats have I got?

 How much can I charge for this?

What's your opinion on

how much of say a hundred seats is gonna fall into the most extreme version of I must

get this right? I want better term. I wonder what you think the breadth of that looks like. Do you have a sense of,

is this 50 50, is this like 80% of cases people are happy to automate?

 What does your gut tell you? 

James Ollerenshaw: I think this is where you start to see

outcome-based pricing, usage based

pricing compared to seat based pricing becoming a consideration. And so in

a situation where

you're still gonna need a fair amount of human interaction along the process, seats can still make sense.

If that's going away, we need different ways of pricing it. It's gonna depend

what your software does and the markets selling into, and

some all the other effects

we've been talking about. But you're right, 

it's gonna look different for different types of SaaS companies.

Rob Curtis: Ken, what do you think?

Ken Lempit: There's been talk about, if

you have API access and you're revealing that to agents, your pricing power goes away. I think the opposite.

If you're accessing me by API, I

know every time you request work.

I can build pricing against that. 'cause

I know what work I'm doing for you and what value you should

be getting as a transfer, an expectation of value. So it's gonna

take some work because customers are not used to paying much for an API call.

Sometimes 

a platform will have an infinitesimal, API call charge.

But once API calls our unit of work. Then we have to start charging

for a unit of work. Pricing is up in the air. I think it's gonna be

one of the big consequences and probably margin up in the air.

Being a broker

is less of a value proposition

than owning the transaction cycle. You brought up OpenTable. They own the transaction cycle, almost everything

they process, they're the beginning,

middle, and end of the transaction.

If they become just a broker and don't own the

audience, they lose the first party data, they lose a lot of things

in being relegated to the broker role.

So I think it's gonna depend on 

your business model.

On the other hand, for OpenTable,

there's cost of audience acquisition, if they can focus on being the restaurant service provider, maybe that's a more efficient business model and they'll benefit.

Part of

the SWOT analysis is, some

of your top line might

be threatened as

a result of being relegated to a

broker role, but maybe it's

less expensive to be in business as a result, especially if

you're still that important 

to the restaurant.

OpenTable, Toast, all these different things. They do a lot more than just the booking depending on which vendor we're talking about. So maybe they get a cleaner value prop. 

It's a big depends answer.

Rob Curtis: It's

really interesting,

like I don't love any

company that has an API called buy another platform that I care about.

It's really difficult to love a company that provides APIs into a company you engage with day to day.

And my hunch is that this is going to change 

the way in which we think about how to compete

your dashboards doesn't matter.

User experience probably doesn't matter. These big domains of differentiation 

I think are going to start

to push folks that

fall into the invisible software category, 

into commoditization. What I like

about OpenTable is

it sends me a little

WhatsApp and 

compared to the restaurant, I can

achieve the same

functional thing. But because they own the customer relationship,

they have more ways to surprise and delight.

And I think companies that fall into the realms of just being an 

API provider, need to have a hard look at

their business and say, do I need this 

design team? Do I need this user experience team ? Do I need a marketplace of restaurants at all? Or do I go to transaction processing for your restaurant bill?

And I think that just

occupies an incredibly different space

in the mind of consumer. And for a new software provider

like myself, that's a great opportunity, which

is how can

I provide that 

API call at a cheaper price.

In the commoditization

world, if you're invisible software, there's no customer

loyalty or love, then the agent

is probably going to be operating on instructions, like, find me

the cheapest way of doing this thing.

These

are the real questions that many SaaS providers are going to want 

to be asking themselves

as they start to see for their particular industry and use case, what is happening to players

like me and

how do I compete in a world where

most cannot afford the investment in owning the

agentic player?

It's complex, it's multi-step, it's gonna be a real battle. And

if you're falling

into the realms of invisible software, it's going to be how do you protect your margin by drastically reducing

the scope of what value looks like for your customers. These are gonna be interesting

things to pull

on as we

see more

real examples hit the market.

 Ken, I'm gonna hand over to you.

This has been a great discussion, but as always, we're onto predictions, so I'll let

you take it from here.

Ken Lempit: I think I have to go back to the prediction

around what is an agent, what is a co-worker? 

So I'm gonna say, the

co-worker is king.

The idea that I can engage with and

benefit from a relationship with

a digital co-worker, 

is where there's a lot of opportunity.

And I think this idea of calibration of that skill level

also is something that

Might actually happen. That's my prediction, calibration of AI effectiveness.

Rob Curtis: James predictions for this week.

James Ollerenshaw: When you look at the

history of technological revolutions they are rarely binary. 

We don't have a clean before and after. 

We end up with a mixture some of the old, gets pushed out but often doesn't completely go away

as the new comes in.

 I think it's gonna be much the same with

this era of AI and Agentic AI. 

There's gonna be many instances where we are very happy to 

the AI take care of it. 

Sometimes we're not. Sometimes we wanted to do it for ourselves. And in between there, there'll be areas of grey where we wanna be

able to be involved with the computer.

And I wanna go back to the dishwasher,

How many times have you thought, I'd love to see what's going on inside the dishwasher? 

What, why can't the dishwash have a glass door?

There's some reasons

why it doesn't. That's not important.

We wanna see what's going on

inside the machine and how things are

working in many cases.

And so, sometimes, yeah, we can set it and forget it. Sometimes we'll wanna

do it ourselves and sometimes it's gonna be in between.

Rob Curtis: What was your prediction?

James Ollerenshaw: Agents are coming but it's gonna be non-binary. We'll see a great expansion in the use of

those, but it will not be total.

Rob Curtis: Yeah, 

that feels right.

And there's a lot of, it depends and what's your

James Ollerenshaw: Yeah, 

Those dependencies are really important. 

If you are a software company what is your niche?

What is your place in 

that broad ecosystem? You might

want to take a narrow part 

and focus on that,

 But for, the broader business and economy, 

we'll see a greater shift into AI but some of the old's gonna stick around too.

Rob Curtis: Thank you James. last week

I predicted agent

optimization is going to be

the future of discoverability. This week I

wanna talk about voice and we're gonna have a whole show on voice next week.

I think voice is going to

drive the agentic revolution across a majority of

consumer use cases, and 

start getting into SaaS. 

What does that mean? 

Agents will begin to be optimized

for tasks that can be described simply in a short conversation. I write prompts

that are this big right

now, but I suspect as we get

closer and closer to being able to speak to computers,

we are going to then

find ourselves influenced in terms of the things

that we can ask for and how we engage with them.

I'm gonna predict that

there is going to

be a large

aha moment around voice this year in 2025. 

I'm gonna hold myself to a very specific prediction here,

and I'm also going to predict that it

will not be Siri. Some

new player is going to come out and

change the game in terms 

of how consumers and buyers

and people across the world think about the role of voice in AI.

And I think Siri is gonna miss the boat on this one. 

I think it's gonna be a new

player that's gonna come out with a unique twist on how to use voice plus Agentic AI to solve 

real meaningful problems for consumers. 

Ken Lempit: I want another bite at the apple now. 

 So building on this vision you have for voice,

I think it's also

gonna be the

year where it's a conversation

where your AI will remember what you talked about previously. 

Right now when you engage by

voice or keyboard, it's like you're

starting over. It's like Groundhog Day, right? Every interaction is almost

a new thing. Unless it's right in the thread, you can't go to

ChatGPT and say, Hey, remember last Wednesday we talked about, building a

three story apartment building?

It's like,

sorry. So that, I think

continuity of conversation is gonna become

really important this year, and

that'll be a breakout. 

Rob Curtis: You'll be pleased

to see OpenAI pushed out a major memory enhancement that will make that true

as of yesterday.

 Memory's been

a big constraint and

OpenAI pushed out an update late

last night which will be, 10th of April for 

listeners. It can now remember many of your

conversations first time round.

I've not tried it yet

'cause I'm also finding an exhausting to have to say the same things. 

But we're seeing big

developments in memory

that should make the utility increase a hundred fold.

Ken Lempit: This has been

a really great conversation.

For those of you loyal listeners

to the SaaS backwards podcast, you'll recognize this is at least twice as long as the usual

episode, even after we've edited it down. For that, I apologize,

but I hope you'll also

be appreciative of the depth of this conversation.

Rob curtis, building applications in his venture studio.

Standup Hiro is about to come out and I

think by the time this is published, 

Standup Hiro will be available.

And that's HIRO

James Ollerenshaw, technology marketing

executive AI guy multiple

times over, AI marketing lead multiple times

over. 

Thank you so much. If you haven't subscribed already to the podcast, please do so wherever podcasts are distributed, and also you can get the podcast on

YouTube. Just search SaaS backwards.

Hey rob, if people wanna reach you, how

can they do that,

Rob Curtis: I encourage everyone to

go to www.standuphiro.com. There are plenty of ways

to get ahold of you from there to speak to

some of our agents. And you can also reach me on LinkedIn. Rob curtis. 

Ken Lempit: My Demand Generation Agency for software as a service companies is Austin Lawrence.

You can reach us

at austinlawrence.com

I'm on Linkedin/in/kenlempit and thanks everyone for

staying with us and we'll see you

next time when we get deep into voice 

Agentic AI.

I.