In February of this year, I spoke with Ted Elliott, CEO of Copado, about the role that agentic AI and Agentforce are likely to play in the Salesforce ecosystem. It’s fair to say he didn’t pull any punches when it came to the reality of the situation. “Agentforce is really going to kill some jobs,” Ted explained. “Salesforce professionals will inevitably be affected by how Salesforce are going to use AI in the future. Entry-level jobs, [like] junior developers, and admins, might be replaced by agents.” Fast forward to now, and the discourse around agentic AI’s place in modern workforces is starting to take shape.
Outside the Salesforce ecosystem, we’re already seeing extensive cases of agents at work – Klarna’s AI assistant now reportedly handles the workload of 700 support agents, while GitHub’s Copilot Workspace can plan, write, and debug code with minimal human intervention. In law and finance, tools like Harvey AI and IndexGPT are taking over research and drafting tasks, signaling a broader shift where AI agents are becoming integral team members rather than just tools.
Returning to the ecosystem, many are already aware of the looming presence of Agentforce, which Salesforce are continually strengthening and updating rapidly as they seek to create a seamless agent experience for users and customers.
And at Salesforce-focused AI automation company TestZeus, we’re already seeing this transformative shift take place. According to the company’s CEO, Robin Gupta, AI agents won’t just assist with Salesforce work but will be able to take on much of the effort themselves. Tasks traditionally allocated to junior team members – from testing to sales prospecting – are already being automated at scale.
I had the pleasure of meeting Robin for the first time at the London World Tour this year and had the opportunity to sit down with him and delve deeper into the exciting future of TestZeus and agent workforces.
What Is TestZeus?
Before we discuss the “elephant in the room,” it’s important to outline Robin’s background and what he is driving at TestZeus.
Having stepped away from his role as VP of Product and Technology at Provar last August, Robin launched TestZeus, which was inspired by his long-standing interest in autonomous computing.
“For a long time, I was fascinated by the idea of a computer running on its own – opening a browser, clicking buttons, all without human input,” Robin explained. “But doing that has always required technical skills; you have to know how to code. Even with so-called low-code or no-code tools, they only stay that way while the sales rep is on the call. As soon as you’re left alone, you’re faced with a mountain of code to automate a browser.”
In essence, Robin saw through the marketing of “no-code” tools and realized that test automation still requires deep coding skills. Having repeatedly faced this challenge throughout his career, Robin believed that the best time to solve this challenge was in the dawn of the agentic AI era.
“Test automation has remained both highly technical and expensive. Those were the two main problems I kept encountering across different organizations. With the rise of AI and agents, I felt the time had come to fix this. So I resigned, pulled together my research, and began building an AI agent specifically for Salesforce testing and automation.”
“Jobs Will Be Replaced, Not Displaced”
When it comes to the discussion around AI job replacement, the common theme that I’ve found is that the recently graduated entry-level candidates are the ones who should be most concerned. If we’re to look outside the ecosystem again, CEOs from Amazon, IBM, Salesforce, JPMorgan Chase, and BT Group are openly warning that AI will transform or eliminate many corporate jobs, especially in customer service, software, HR, and development.
We recently covered on Salesforce Ben that within the ecosystem, it’s entry-level admins that could potentially be at risk. As AI becomes more sophisticated, it will be able to perform “basic” admin tasks through natural language prompts, ultimately removing the need for human intervention.
When I asked Robin about this, he expressed that we’re more likely to see jobs transformed rather than jobs replaced.
“I’m an optimist, so let’s keep this positive! I believe jobs will be replaced, not displaced, meaning the total number of jobs may change, but the overall impact will still be positive,” Robin explained. “Say we had 100 jobs before – maybe 99 remain, but that one person who’s ‘replaced’ ends up doing something new. That’s the shift we’ll see.”
In terms of people ending up “doing something new”, Robin practically described it as “free promotions”. In essence, agentic AI will most likely act as an enabler for employees to level up faster than ever before.
“Society as a whole will level up. Everyone gets a kind of promotion. For example, I used to be a manual software tester. Thanks to tools like ChatGPT, TestZeus, and Hercules, I can now operate like an automation expert without needing to spend eight months learning scripting.
“So, in my mind, everyone gets a free upgrade. I’m a terrible illustrator, but with generative tools, I can now design well enough in just three days. That’s a design promotion I didn’t expect.”
On top of this, Robin digressed that AI agents still need human oversight. So, while there may be an emerging agent workforce, we still need people with deep knowledge monitoring the performance of agents.
“Every sweet thing needs a pinch of salt. The practical side of this is what Salesforce describes as the digital labor model: every AI system still needs a human operator. Whether it’s ChatGPT, Agentforce, or TestZeus, someone still needs to press a button and check the output.
“True displacement only happens when no one needs to review the results anymore, and we’re still a long way from that.”
“Agentforce Is Like a Very Fancy Cadillac Car…”
Agentforce is Salesforce’s most ambitious product to date, and it feels as though the CRM giant are putting all their eggs in one basket when it comes to the AI tool.
But the current sentiment in the ecosystem is that adoption has been hindered by unclear pricing, steep implementation requirements, and a workforce that needs additional support before adopting it (which Salesforce are actively working on at the moment).
Robin echoed many of these points when it comes to Agentforce – while exciting, it’s not in a position yet to replace workforces.
“See, Agentforce is like a really fancy Cadillac. Everyone’s enamored by it – ‘Oh, I want a Cadillac too.’ But the moment I tell you the price, the features, and the mileage it gets, suddenly it’s, ‘Hmm, maybe next year, not this year.’”
Robin also pointed out a current skills gap that is causing adoption issues, arguing that Agentforce is currently being marketed as a “plug-and-play” AI, when the reality is that a deep understanding of prompt engineering, LLM configurations, and data structure alignment is still required.
“The devil’s in the details. Everyone wants an agent that can handle support or sales, and now there’s even an employee agent, which is more general-purpose. But there are two big problems I’ve seen and heard from others in the community.”
“First, the pricing is complicated and has already changed multiple times. By the time you go from a POC to a POV to actually raising an invoice, the pricing may have changed twice. That makes buyers hesitant. They’re saying, ‘I don’t trust this yet. I don’t know how my credits will be used. What’s a Data Cloud credit versus a Flex credit?’ So that’s one pain point.
“Second, the users themselves – such as developers, BAs, testers – are often operating at what I’d call a Level 2 understanding of large language models and RAG (retrieval-augmented generation). But Agentforce is being marketed at Level 99 – like you can just plug in Data Cloud and RAG and it’ll magically work.
“In reality, developers don’t know what a proper RAG setup looks like or how to write a solid prompt. There are some guidelines now, but if you check the Agentforce Slack channel, you’ll see people struggling. ‘It’s not giving me the right SOQL query.’ ‘My database says I have this many orders, but the agent says something else.’
“The users are behind the product, while the features keep racing ahead. You’ve gone from Einstein to Agentforce to Agentforce 2, and now Agentforce 3 is already on the horizon.”
So, while the “Agentforce vision” comes to fruition down the line, there are still a number of roadblocks that users have to get through before we see agents displacing roles at a large scale across Salesforce.
“The Majority of Our Work Will Be Done by AI Agents”
Robin’s prediction on jobs being displaced rather than replaced acts as a reassuring look at what the future of most workforces may look like, but when it comes to the structural future of TestZeus, the vision is simple – keep the team small and have a network of agents completing tasks.
“I can’t speak for everyone, but at TestZeus, we strongly believe in building a company that stays small in size, with the majority of our work handled by AI agents. Even today, much of our sales prospecting, emails, follow-ups, and voice calls are done by agents because we don’t have a lot of people, and we don’t intend to.
“I believe that anything which can be done on a phone, behind a screen, or with a keyboard will be done by an agent within the next two to three years. And when you think about it, that’s a huge portion of today’s jobs.”
Robin also explained that this is something we are likely to see at scale in the Salesforce ecosystem, arguing that higher quality is and will continue to come through, and it will enable small teams to match enterprise-grade output.
“In the Salesforce ecosystem, there are generally three types of service providers. At the top, you have the big players like Accenture, TCS, and others. Then there are mid-sized companies with 500 to 5,000 employees. But here’s the interesting part: 60 to 70% of Salesforce implementation teams are actually very small, just one to five or one to ten people.”
“So you can think of it as a pyramid: a few giants at the top, a smaller group in the middle, and a massive base of small consultancies and freelancers. What we’re starting to see, and what we expect to accelerate, is that the bottom and middle tiers will start performing at the same level as the top tier because they’ll have AI agents doing the heavy lifting.
“They won’t need to keep extra humans ‘on the bench’ to scale up delivery. With agents in the mix, small teams can punch above their weight, and in some cases, even outperform the larger firms.”
And ultimately, the biggest driving factor for replacing human workforces with AI is costs and salaries. This has already been identified as the biggest factor for many companies – Klarna, for example, have been detailing over recent months how much money they’ve saved from shrinking their headcount (despite recently backtracking).
The truth for companies like TestZeus is that if you can get agents performing to their full potential, then the cost-saving aspect becomes very hard to ignore.
“There’s one invigorating question I often ask myself,” Robin said. “What is the salary threshold?”
“For a business leader, whether it’s someone like you, me, or a VP at a firm, it ultimately comes down to this: How much am I paying for a person’s seat, and how much work are they actually getting done?
“In the US, the median QA salary is around $90K. Now imagine I come in and say, ‘We can do the same work for $20K.’ My agents don’t sleep, they don’t take breaks, and they can deliver four times the output.
“At that point, it’s really just about quality and economics. We’ve been offshoring for decades, my first job was as an offshore QA for a US bank. So the real question is: When does that salary threshold finally get broken by AI?”
The “salary threshold” is essentially the point where AI agents become the obvious business choice because they deliver better output for less money. Once that happens, he believes widespread adoption is inevitable.
Final Thoughts
While Robin is highly optimistic and enthusiastic about agentic workforces, it’s clear he believes many humans will benefit from it, whether that’s through “free promotions” or the ability to skip years of technical upskilling and immediately operate at a more strategic level.
His vision isn’t one where humans are replaced entirely, but rather one where agents handle the repetitive, time-consuming tasks, allowing people to focus on creativity, judgment, and higher-value work.
But as we’ve seen with tools like Agentforce, there’s still a long road ahead, marked by pricing confusion, implementation gaps, and a workforce that’s not fully equipped to harness this technology yet. Navigating that path with honesty, pragmatism, and a focus on real human outcomes will be what separates the success stories from the cautionary tales.
