Inside the 'Prove AI Can't First' Company
The Human Justification Report: Your Job is Now on Trial
Welcome to the new reality of corporate hiring: prove AI can't do the job first.
Leaders like Shopify’s CEO Tobi Lütke are already voicing this "AI-first" philosophy, issuing directives that managers must first prove why AI cannot accomplish a task before requesting more human headcount. This move, communicated in internal memos and then shared publicly, reframes AI not as an optional tool, but as a baseline expectation. It forces a cold, hard question into every hiring discussion: When, exactly, is a person worth more than a bot?
Today we’ll dive into:
Inside the policy changes - where they come from, where they’re going
A new role for people managers - ‘Automation Auditors’
The future of Performance Reviews - ‘Job Trials’
Let’s go!
The C-Suite Calculus
In April, Shopify CEO Tobi Lütke issued a directive to employees and managers that before requesting additional headcount, they must prove why AI cannot accomplish the work. This policy was communicated in an internal memo and publicly shared, emphasizing that AI usage is now a baseline expectation and a skill that must be learned and integrated and will also be incorporated into performance reviews.
This new corporate mandate isn't emerging in a vacuum. For nearly two decades, the tech industry was fueled by an era of near-zero interest rates, creating a "growth at all costs" culture where chasing market share and user numbers often overshadowed profitability. That era is definitively over. Now, buffeted by a new economic reality and intense shareholder pressure, tech is living through its "efficiency years." The new mantra, echoed in boardrooms from Silicon Valley to Stockholm, is about "doing more with less," maximizing productivity, and proving a clear path to profit—and AI is being positioned as the ultimate tool to enforce this new discipline.
At its core, the policy is controversial because it can function as a financial sleight of hand. Companies are currently funneling billions into AI—investing in massive computational power and, as Shopify’s CFO Jeff Hoffmeister noted, hiring "a higher comp, high-end AI engineer" whose salary alone can raise costs. This massive capital expenditure creates immense pressure to show immediate ROI. The easiest place to find those savings? Freezing or cutting human headcount in other areas. This creates a challenging reality where roles are being eliminated to pay for the AI bet, often before the technology is mature enough to fully handle the displaced work, leading to potential gaps in quality and capability.
The real-world impact is clear: Shopify’s headcount has been kept relatively flat after significant workforce reductions in previous years (headcount fell to 8,100 at the end of December from 8,300 a year earlier, according to its latest annual filing. The Canadian company eliminated 14% of its workforce in 2022 and 20% the following year), a direct consequence of this new operational discipline.
“What would this area look like if autonomous AI agents were already part of the team? This question can lead to really fun discussions and projects.”
That’s a great question. A better question however, will be “What can AI actually do” and “Where should there be a Human Premium”? We don’t know what will happen in 5 or 10 years, but the tale from Klarna can help us figure out the answer for today’s world. After boasting that its AI chatbot was doing the work of 700 agents, the company reportedly had to resume hiring humans due to quality issues.
Managers: the new AI Auditors
Following Lütke’s directive to its logical conclusion, before a manager can post a job, they must first produce what amounts to a 'Human Justification Report.' It should answer questions like:
Can an LLM draft these reports?
Can a chatbot handle these client inquiries?
Can an AI analytics tool perform this data analysis?
But then go on a deeper quantitative level, because yes, an LLM could draft the reports, but in which job can you currently give the keys to the LLM and then forget about it?
Managers are going to be doing a lot more process mapping and AI heatmaps:
and new process designing, to understand where there is a human in the loop or not:
I’m starting to think of it like ‘Human-AI Engineering’ where managers will need a mix of soft skills and evaluation capabilities, but also harder skills of process mapping and AI measurement (another interesting topic we’ll have to discuss soon).
As often the case, while CEOs set the strategic vision, there’s a ton of nuances and complications underneath the surface. For example: how do you quantify the ROI of a team member who excels at mentoring junior staff, de-escalating a tense client situation, or having a spontaneous conversation that sparks a million-dollar idea? These uniquely human contributions are too difficult to capture on a spreadsheet comparing a human salary to an API call.
So, this new burden will likely create two outcomes. First, managers will learn to game the system. They'll become experts at highlighting the unquantifiable, framing tasks in ways that emphasize ambiguity and ethical nuance, effectively fighting a bureaucratic battle to get the human help they know they need. Second, it could create a chilling effect on innovation. If any new project or responsibility is immediately subject to an "automation audit," will managers be less likely to experiment or expand their team's scope, opting instead to stick to the already-justified "human" tasks?
There is also a more optimistic—though challenging—potential outcome. If executed thoughtfully, this mandate could paradoxically bolster innovation by removing traditional skill barriers. When AI can handle baseline technical execution, it should allow employees to be more eclectic and ambitious; a brilliant product strategist who can't code could suddenly prototype an app, or a data-savvy marketer without design skills could generate compelling ad creatives. This democratizes creation, shifting the focus from "do you have the technical skill?" to "can you orchestrate the tools to create value?" But this positive path hinges entirely on implementation. The outcome depends on how the policy is communicated (as an empowerment tool or a cost-cutting threat), enforced (as a flexible framework or a rigid checklist), and incentivized (rewarding novel outcomes over mere headcount reduction). The answers will determine whether this policy creates a culture of fear or one of unprecedented, broad-based innovation.
AI Performance Reviews & Employee Job Trials
The mandate from leaders like Tobi Lütke to "Prove AI Can't First" isn't just about hiring freezes; it fundamentally changes the social contract with existing employees. Lütke made it clear that at Shopify, AI usage will be incorporated into performance reviews. This cements AI proficiency not just as a desirable skill, but as a core, non-negotiable component of job performance itself. The annual review is evolving into a continuous job trial, where your value is constantly measured against the ever-growing capabilities of AI.
On the surface, the logic seems clear. We can envision a new performance dashboard where employees are measured against a fresh set of AI-driven metrics:
Automation Quotient: What percentage of your manual, repeatable tasks have you successfully automated this quarter?
Tech Stack Fluency: Which AI systems (from internal LLMs to external SaaS tools) did you leverage to achieve your goals?
Adaptive Learning: Demonstrate proficiency in at least one new generative AI model or platform that was released in the last six months.
Efficiency Dividend: Show the quantifiable time and cost savings your new, AI-augmented workflows have produced for the company.
This is the new "Job Trial." The implicit question being asked is no longer just "Did you do your job well?" but "Did you make your job more efficient and valuable than an AI could have?" But this approach, while sounding like a data-driven utopia to a CFO, is fraught with peril. It risks creating a "quantification trap," where employees are incentivized to automate the easiest tasks, not necessarily the right ones. It can perversely reward employees for becoming expert "AI tool jockeys"—constantly learning new systems—rather than engaging in the deep, strategic, and creative work that truly drives innovation.
The fundamental problem is that this new, complex layer of measurement is being built upon an already broken foundation. Let's be honest: the traditional performance review is one of the most widely criticized rituals in corporate life. Many companies do them poorly or only once a year, they are notoriously subjective, and they consistently struggle to assess the true value and merit of an individual's contribution—especially the "soft skills" of collaboration and mentorship.
Adding a slate of easily-gamed AI metrics to this flawed system is like installing a sophisticated navigation system on a car with a broken engine. It won't fix the underlying problem of how to measure human value; it will just create more ways for the system to be inaccurate, demotivating, and distrusted. This could easily devolve into a system that measures activity over impact, fostering the very culture of fear and mistrust that innovative companies should be trying to dismantle.
Conclusions
Whilst I appreciate that especially in large organizations more ‘questioning’ is very valuable, the risk is that this kind of approach will lead by fear instead of enthusiasm. It will also create new kinds of tensions and mistrust between employers and employees - a ‘metric’ that is at the core of success of an organization.
I get it—in large organizations, more "questioning" can be valuable. But there's a huge risk that this approach will lead by fear instead of enthusiasm, creating new tensions between employers and employees.
I’m also seeing more "AI Organization charts" where bots get job titles. As a marketer, I can tell you this is a pure marketing stunt. SaaS companies are rebranding their tools as ‘digital workers’ because it’s much easier to justify a $30,000 annual fee for a "worker" than for a piece of software. It’s a toxic framing that benefits no one.
So here’s my $1M question: where is HR in all of this?
The answer is that HR is facing a choice: become the bureaucratic enforcer of this new "AI-first" mandate, or step up to become the architect of a far more critical discipline: Work Design. If HR is just the department that double-checks the manager's 'Human Justification Report,' then it’s already failed. Its real job is to preempt the report altogether by redesigning the work itself.
This isn’t about drawing new process maps. This is about asking smarter questions than the C-Suite: not can AI do this task, but how can AI handle the 80% of grunt work to free up a human for the 20% that actually drives value—the client schmoozing, the creative leap, the ethical gut check. This approach shifts the entire game from a defensive justification of human existence to an offensive strategy of human amplification.
This human-centric work design creates the conditions for a new archetype to emerge: the super worker. Let’s be clear: this isn't just someone who knows the right prompts. The super worker orchestrates a suite of AI tools as an extension of their own expertise, achieving outcomes that were previously unimaginable. They are the new power users, the living embodiment of the "Human Premium."
By cultivating these individuals, HR can offer the C-Suite a far more compelling ROI than the short-sighted savings from a hiring freeze. The goal evolves from trimming headcount to amplifying talent.
Ultimately, the path of "Prove AI Can't First" is a managerial dead end. It fosters a culture of fear, mistrust, and metric-gaming that is the antithesis of innovation. The real future isn't about creating AI org charts or justifying salaries against API calls. It's about building a company full of super workers. The only question is whether HR will lead that transformation or just be left holding the clipboard.
Ciao,
Matteo