Beyond the AI (or Agent) First Hype: Why Human-Centered AI Strategy Actually Works

A year of rehiring, refunds and reality checks shows why putting humans at the center still delivers the results “AI first” promised.

Every startup pitch deck seems to lead with it. Every enterprise transformation strategy mentions it. In 2026 the slogan has a new sequel: “agent first.” But the uncomfortable question no one’s asking hasn’t changed: AI first for whom?

Since I first wrote this piece, the evidence has piled up on both sides. Some companies have quietly rehired the people they replaced. Others are finally seeing real returns. The difference between them is almost never the model. It’s how humans were placed in the picture.

The AI First Illusion

Walk through any tech conference today and you’ll hear endless variations of the same promise. Last year it was “AI-powered.” This year it’s “autonomous agents” that will run your workflows end to end. Labels get slapped on everything from coffee machines to calendar apps.

Much of it is relabeling. Gartner calls the practice “agent washing”: vendors rebranding chatbots, assistants and RPA tools as agents. It estimates only about 130 of the thousands of vendors claiming agentic AI offer the real thing (Gartner).

AI doesn’t want to be first. AI doesn’t want anything at all.

The companies shouting loudest about being “AI first” often reveal how shallow their understanding is. AI is a sophisticated tool that amplifies human capability. It is not a replacement for human judgment, creativity and strategic thinking, and agents don’t change that. They just raise the stakes when the judgment is missing.

The Human Reality Check

Here’s what the “AI first” evangelists don’t want to admit: without humans willing and able to make AI work, there is no “AI first.” There’s just expensive software sitting unused in digital warehouses.

The past year turned that from opinion into data. A Careerminds survey of 600 HR professionals in February 2026 found that more than half of those who made AI-driven layoffs said AI needed more human insight than expected. Only about 21% said AI fully replaced the roles with no operational problems (HRD).

Every successful AI implementation I’ve witnessed has one thing in common. It started with humans who understood both the problem and the tool. They didn’t begin with AI and work backward to find problems. They started with real business challenges and discovered where AI could genuinely help.

The Right Questions to Ask

Instead of asking “How can we be AI first?” successful organizations ask:

Where can AI actually help? Not everywhere. Not automatically. But in specific, well-defined scenarios where pattern recognition, speed or computational scale creates genuine advantage.
What problems are we actually solving? Real problems with measurable impact. Not theoretical improvements or solutions in search of problems.
How is AI solving these specific problems? Through clear mechanisms that teams can implement, monitor and improve over time.
Does this even need an agent? Gartner notes that many use cases marketed as agentic don’t require agentic implementations. A simpler assistant or plain automation is often cheaper and safer.
Who is accountable when the AI is wrong? Every output a customer, regulator or client relies on needs a named human who checked it.
What are the real impacts? Quantifiable outcomes: reduced costs, improved quality, faster processes, better customer experiences. Not just “AI magic.”
What resources do we actually need? Human expertise, data quality, infrastructure, ongoing maintenance. The unglamorous foundation that makes AI tools effective.

Beyond the Hype: What Actually Works

The companies creating lasting value with AI share a common approach. They’re human-centered, with AI enablement.

McKinsey’s latest State of AI survey (November 2025, nearly 2,000 respondents) puts numbers on this. Most organizations now use AI somewhere, yet only 39% report any enterprise-level EBIT impact, and for most of those it’s under 5%. A small group, about 6% of respondents, qualify as “high performers” (McKinsey).

What sets them apart isn’t better models. High performers are nearly three times as likely to fundamentally redesign workflows around AI, rather than bolting it onto old processes. They’re three times as likely to say senior leaders visibly own the AI effort. And they aim for growth and innovation, not just cost cutting.

In other words, they start with human expertise and judgment. They identify where AI can amplify that capability. They measure success in human terms: better outcomes for customers, more engaging work for employees, stronger results for stakeholders.

AI becomes powerful when it makes humans more capable, not when it tries to replace them.

The Cost of Getting This Wrong: Real Failure Cases

Organizations that chase “AI first” without a human-centered strategy end up with technology that doesn’t fit real workflows, solutions to problems no one had, teams that don’t trust the systems, and metrics that look good in slides but not in the P&L. The evidence keeps mounting.

The AI layoff regret wave. This is the biggest story of 2026. Forrester’s Predictions 2026 report says 55% of employers who cut staff citing AI now regret it. Gartner forecasts that half of companies that cut customer service headcount because of AI will need to rehire for similar roles by 2027 (People Matters). Careerminds found about a third of companies with AI-led layoffs had already rehired 25–50% of the roles they cut.

Klarna: from AI first to hybrid. Klarna once said its AI assistant did the work of 700 agents. In 2025 its CEO admitted the cost-driven push had produced lower-quality support and promised customers they could always reach a human (CXM). By early 2026 Klarna was recruiting its own loyal customers as part-time, Uber-style service reps, with AI handling the basics and humans handling the moments that matter (CX Foundation). Klarna didn’t abandon AI. It put humans back at the center.

Deloitte’s hallucinated government report. In 2025 Deloitte Australia agreed to partially refund a AU$440,000 government contract. Its 237-page report contained a fabricated quote from a court judgment and references to research that doesn’t exist. The revised version disclosed that Azure OpenAI GPT-4o had been used in producing it (AP). The tool wasn’t the failure. The missing human check was.

Microsoft Copilot: wide access, thin adoption. Microsoft disclosed 15 million paid Microsoft 365 Copilot seats in January 2026. That’s real growth, but only about 3.3% of its roughly 450 million commercial Microsoft 365 users (Directions on Microsoft). Analysts point to confusion, cost and hard-to-prove value. Access was never the bottleneck; change management was.

McDonald’s AI drive-thru. After about three years of testing with IBM, McDonald’s ended its AI drive-thru experiment in 2024 (CIO). Viral videos showed the system adding unwanted items, including one order that ballooned to 260 chicken nuggets while customers begged it to stop. What worked in the lab failed at the speaker box.

IBM Watson Health: the original cautionary tale. IBM spent more than $4 billion on acquisitions to build Watson Health. Watson for Oncology was reported to make unsafe or incorrect treatment recommendations, in part because it was trained on hypothetical cases rather than real patient data. In 2022 IBM sold the unit’s data and analytics assets to Francisco Partners, reportedly for around $1 billion; they now operate as Merative (Healthcare IT Today).

The agent reckoning ahead. Gartner predicts more than 40% of agentic AI projects will be canceled by the end of 2027, citing rising costs, unclear business value and weak risk controls (Gartner).

A note on the famous “95%” stat. MIT NANDA’s 2025 report claimed 95% of generative AI pilots deliver no measurable P&L impact. It went viral, but critics have since pointed out its limits: a small sample (52 interviews, 153 survey responses), self-reported results and a roughly six-month window (Marketing AI Institute). Treat it as a warning sign, not a universal failure rate. Its core diagnosis still holds: pilots stall because of a learning and integration gap, not because the models are weak.

The common threads haven’t changed: overpromising, underestimating implementation, ignoring users and putting technology ahead of human workflows. Meanwhile, human-centered competitors are quietly using AI to serve customers better and build advantages that last.

The Path Forward

The future belongs to organizations that understand a simple truth: AI is not a strategy. AI is a capability that enables better strategies.

The most effective approach isn’t “AI first,” “agent first” or even “human first.” It’s human-centered AI strategy. That means:

Starting with clear human needs and business objectives
Understanding where AI capabilities genuinely align with those needs, and where a simpler tool will do
Redesigning the workflow, not just dropping AI into the old one
Keeping a named human accountable for anything customers, clients or regulators rely on
Measuring success through human and business outcomes, not seats sold or pilots launched
Treating workforce cuts as the last lever, not the first, and keeping the people who know how the work really gets done
Continuously improving the human-AI collaboration

The Bottom Line

Hype doesn’t win games. Results do.

The last year proved it. The companies rehiring, refunding and rethinking are the ones that put AI first. The ones pulling ahead redesigned their work around people and used AI to make those people better.

AI without humans taking constructive action is just expensive software. Agents without human oversight are expensive software that can act on their own mistakes. But humans empowered with the right AI tools? That’s where the real transformation happens.

At H4Tech, we’re excited by how much value people can deliver when technology works in their service. Every advance we adopt has one job: help our teams deliver efficiently, reliably and faster — not just faster. No hype as a service. Just what humans can deliver, amplified by technology.

Sources and References

People Matters — 55% of employers regret AI layoffs; Gartner predicts more rehires by 2027
HRD — Businesses rush to rehire staff after regretted AI-driven cuts
CXM — Klarna rethinks AI push, rehires humans after automation stumbles
CX Foundation — Klarna recruits its own customers as service reps
Forbes — Klarna reverses on AI, says customers like talking to people
AP — Deloitte to partially refund Australian government for report with apparent AI-generated errors
Directions on Microsoft — Microsoft claims 15 million paid M365 Copilot seats
No Jitter — 4 obstacles impede paid Microsoft 365 Copilot adoption
CIO — Famous AI disasters
Healthcare IT Today — IBM Watson Health sale to Francisco Partners
IEEE Spectrum — How IBM Watson overpromised and underdelivered on AI health care
Gartner — Over 40% of agentic AI projects will be canceled by end of 2027
McKinsey — The state of AI in 2025: Agents, innovation, and transformation
Fortune — MIT report: 95% of generative AI pilots at companies are failing
Marketing AI Institute — That viral MIT study claiming 95% of AI pilots fail? Don’t believe the hype

Subscribe