Opinion
The PC productivity paradox, again
Companies risk repeating the mistakes of the PC era by layering AI onto outdated processes. The winners will be those that rebuild their organizations around autonomous workflows.
I remember the absurdity firsthand. The executive who had every email printed and filed it in a physical folder. The financial form you would download, only to be required to fax it to the bank. Technology always arrives faster than organizations can change.
Most companies will adopt agentic AI the same way they first adopted the PC. They’ll hand work to an agent, have a human review every output, copy it into another system, wait for another approval, and only then launch the next agent. That’s printing the email all over again.
Economists spent the late 1980s puzzling over why computers showed up everywhere except in the productivity statistics. The eventual answer was uncomfortable: the technology worked well, but the organizations refused to change.
The gains took many years to arrive, and they arrived only at companies that redesigned how work flowed, not just what tools performed it.
Successful applied AI startups will usher in this agentic transformation by helping businesses reorganize their workflows—or they will seize the opportunity themselves by leveraging agentic workflows to outrun the legacy players.
Most companies will fail the agentic era in the same ways we saw in the PC and internet transitions. This multi-faceted challenge is the startup’s opportunity.
Workflow Redesign
Companies digitized documents only to print, sign, scan, and file them because the workflow never changed. The same mistake is coming with agents: every completed task becomes another item in a human’s inbox instead of triggering the next step automatically.
Engineering organizations are living this already. Coding agents were adopted en masse with no change to process, organization, or infrastructure, and the result is a swelling queue of AI-written code awaiting human review.
The bottleneck didn’t disappear; it just moved downstream. An overflowing pull request queue is the new overflowing inbox.
Decision Rights
The PC gave employees real-time information and analysis, but every meaningful decision still waited for a manager. Information moved at the speed of light; authority moved at the speed of the org chart.
If every low-risk agent action waits for human approval instead of being delegated within clear guardrails, you haven’t deployed agents. You’ve hired very fast interns with no signing authority.
Enterprise Integration
Companies computerized every department and then left humans to carry information between them. Today the walls aren’t just human. They’re decades of legacy and proprietary software systems that were never designed to talk to each other, let alone to an agent.
The SaaS era made this worse in a way few appreciate: it empowered a manager inside each department to swipe a credit card and buy their own software. That worked because SaaS optimized tasks and workflows within the department. Agentic workflows cut *across* departments, which puts them far above that buyer’s pay grade.
Giving every silo its own AI agent while employees orchestrate the handoffs doesn’t make an agentic enterprise.
Rethinking Work
Agentic AI, like the PC before it, demands that every business take a first-principles approach: how would we structure this company if we were starting it today? That is a brutally hard question for an incumbent to answer honestly. But only when you know what’s possible can you plot your next steps accordingly.
The companies that never ask it won’t only fail at AI adoption—they’ll watch agentic startups turn that hesitation into an opening.
Not From Below
Notice what these four failures have in common. Each one crosses functional lines. Redesigning a workflow means redesigning the organization itself to accommodate end-to-end agentic workflows. No employee, however talented, has the authority to redesign and implement cross-functional workflows.
In the PC era, some executives believed they could hand computers to their young employees and magic would follow. The magic just walked out the door, to the organizations that knew how to take it to the next level. Bottoms-up adoption produces bottoms-up gains: individual productivity, local optimization, and not much else.
Transformation is a top-down decision. That’s why forward-thinking CEOs are suddenly in such demand. It’s why the FDE (forward-deployed engineer) model works: outsiders can see and implement what an organization could never do to itself, for structural reasons as well as technical ones.
And it’s why startups with no legacy structure or mindset can build the operating model natively in almost any vertical, while incumbents are still printing, signing, scanning, and filing.
Adam Fisher is a Partner at Bessemer Venture Partners.
