The AI-human collision — balancing automation and accountability in the new workplace

Simon Ellis, chief executive of Jem, the fintech provider formerly known as SmartWage, believes that in­­stead of losing the human touch, AI can actually do “quite literally the opposite” and act as a “total enhancer” to relationships in a company.

Ellis admits he initially thought AI would depersonalise everything, but his view shifted to believing that AI can create more meaningful relationships with the people who matter by freeing up human attention and highlighting when people need support.

Of course, this pivot from doing wage books for retail clients with casual workers to full-stack HR processes delivered via WhatsApp (powered by AI) has secured an $8.4-million first major round of institutional venture capital led by global fintech investor Quona Capital, alongside the University Technology Fund and E4E Africa, and prominent angel investors such as former Old Mutual CEO Iain Williamson.

This capital injection is earmarked to expand Jem’s WhatsApp-based front-line HR platform into a comprehensive, AI-native workforce management suite.

To its credit, Jem eats what it farms and uses an internal AI assistant itself. This agent scans its communication systems, such as Slack and internal messages, to flag when employees need help or important milestones occur (such as birthdays and anniversaries).

For instance, it successfully flagged that Jem’s chief technology officer had worked three weekends in a row without management’s knowledge, and prompted a supportive mental health check-in.

Rather than replacing human contact, Ellis says, the technology is used to “increase the rate at which a manager can make personalised interventions”. However, he does concede that for this model to work, employees and employers have to be comfortable with there being “a ton of surveillance in that workflow”.

Human in the legal loop

As autonomous systems gain direct access to corporate databases, email accounts and payment systems, the traditional boundary between HR and cybersecurity has dissolved. Anna Collard, head of content strategy and chief information security officer adviser at KnowBe4 Africa, explains this.

“In almost every type of organisation we are now managing a hybrid workforce, and half of that workforce doesn’t have a heartbeat.”

But it’s the scale of this challenge that is highly concentrated: Splunk’s 2026 report on issues concerning information security officers reveals that 96% of them now oversee AI governance and risk across their entire organisations. The Retail and Hospitality Information Sharing and Analysis Center’s report shows 71% of security leaders rank AI as their primary point of friction, overtaking ransomware and phishing.

KnowBe4 Africa data shows 64% of South African organisations report that their AI use is unapproved or ungoverned, and 38% of local security leaders state that AI agents are already taking autonomous action in workflows.

If an employee sets up an AI agent with access to customer data, it represents an HR, security, legal – Protection of Personal Information Act (Popia) – and procurement issue simultaneously. By 2030, security and HR executives must collaborate to enforce machine identity management, AI orchestration literacy and runtime governance (anchored in frameworks like the ISO 42001).

The legal and compliance risk then is still very much on the humans and not on the platforms.

Interestingly, this shift is broadening the entry-level pipeline. Employers are increasingly looking beyond computer science degrees, ranking analytical thinking, problem-solving, psychology and behavioural economics as essential qualifications to govern how humans and machine agents interact.

Sending sensitive HR, payroll and financial data through consumer messaging networks raises even more of these security concerns. Jem addresses this via a three-layer authentication protocol: device-level lock (phone biometrics or passcode), a Whats­App application lock, and password-protection on individual documents such as PDF payslips.

Ellis disputes that consumer channels are inherently vulnerable. “There is this false narrative that WhatsApp is insecure. WhatsApp is incredibly secure. It has three billion active users. People’s emails get hacked in the same way that WhatsApps get hacked. There’s no difference.”

Platform dependency and pricing variables

But building on third-party channels introduces structural vulnerability. If Meta decided to change WhatsApp’s core business model, native applications would face severe disruption.

This risk is already visible in Meta’s global pricing adjustments rolling out on 1 October, designed to limit message spam and maintain high user relevance. Although Jem has a direct relationship with Meta as an official solutions provider, the risk of shifting platform policies is a permanent operational variable.

And the same problem is happening with AI vendors such as Anthropic, OpenAI, Microsoft and Google adjusting the pricing models for tokens and usage.

Organisations also risk undermining their technical investments by failing to engage their workforces, as shown by research from the Top Employers Institute:

When employees feel excluded, they rarely raise their hands in town halls; instead, they quietly disengage, double-­check outputs, or adopt shadow AI tools outside official channels … allowing expected business value to erode.

Sandra Botha, lead HR auditor at the Top Employers Institute, says “adoption is not automatic”. “It has to be designed... Most employees have never been asked how AI is affecting their work, and that gap shows up later in ways that are much harder to fix.”

To successfully navigate this collision, HR and security leaders must redesign organisational governance, culture and employee enablement. Transition from a top-down AI implementation model to a citizen developer framework could bridge the trust gap.

Rather than restricting AI development to isolated specialist teams, Vodacom trained employees in HR, finance, legal and compliance to build their own automated solutions using low-code and no-code tools.

The lesson seems to be, though, that the era of running separate point solutions for corporate problems is ending. Companies must transition to unified, mobile-first platforms to reduce admin friction and operational churn.

To prevent silent resistance and shadow AI risks, employees must be included in AI ethics and implementation frameworks. Up-skilling non-technical staff into citizen developers creates immediate financial and operational value.

And then there’s the rise of autonomous workers without a heartbeat – powered by the agentic AI era – that requires human HR and information security leaders to build collaborative governance models to manage digital identity, Popia compliance and agent access.

Yes, AI is bleeding into the human management side of the business as companies speed up service delivery to their staff and shrink to leaner teams. But someone still has to carry the legal risk – though the bots do a really good job of remembering your birthday. DM

This story first appeared in our weekly DM168 newspaper, available countrywide for R35.