Editorial
Guide
Is AI Making Your Organization More Productive or More Dependent?
Key Takeaways
- Enterprises are designing AI agents to pause for human judgment before high-stakes actions.
- Legal, investment, IT and professional services teams use HITL workflows to reduce costly errors.
- The strongest deployments route human reviews through familiar tools such as Slack, Teams and Outlook.
Most people might think that the goal of an AI agent is to remove the human from their work completely. That the direction we are headed in is full autonomy: a system that drafts the contract, sends the proposal, books the appointment, closes the ticket and reports back to a dashboard.
This kind of narrative has become the belief among many and may even be a source of fear and anxiety for many workers.
However, when I look at what enterprises are putting into production, a very different picture emerges. The most sophisticated AI buyers today are not racing to take humans out of the loop. Rather, humans need to be in the loop more than ever, at exactly the right moments, for exactly the right decisions. This is where real enterprise AI deployment is.
From my experience working with hundreds of enterprises (especially in regulated industries) to transform their workflows with AI, I have collected evidence on how different industries are using human-in-the-loop (HITL) architectures to keep AI running.
Legal Services
In legal practices, AI agents can now read contracts at an unbelievable speed. They surface clauses, flag deviations from preferred language and draft recommendations on what to negotiate and what to accept.
With HITL, none of that work touches a client without an attorney's review first via Outlook or Gmail, where they can approve or reject recommendations. The same logic applies to tasks like legal premise extraction from filings, where AI accelerates the first read through of a stack of documents, but practitioners still decide what counts as a premise worth pursuing.
Real Estate & Investment Firms
Investment firms are among the most active builders of agentic workflows. A typical deal review agent pulls data from half a dozen connected systems (listing platforms, internal pipelines, market data sources, document repositories) and assembles a draft profile of the opportunity.
Then, before any business critical action is taken, the agent will pause for an analyst’s review via Slack or Teams, approving or disproving what the model scraped and how it was interpreted. Only after approval from a human will the AI system continue running, sending and extracting information from the firm’s deal pipeline. A fabricated cap rate or a misread rent roll is not a recoverable mistake. Human judgement remains invaluable here, just like the assistance of AI agents.
IT Support & Helpdesks
Internal IT functions have become one of the most reliable HITL deployments this year. AI models propose ticket resolutions based on prior tickets, a knowledge base of internal documentation and the user's history. An L1 agent approves, edits or escalates before the ticket is sent to a reviewer via Slack or Teams and ultimately passed on to the user.
Reviewers rarely start from a blank page anymore, the user gets a faster answer and no automated reply leaves the queue without a human having seen it.
Proposal Generation in Professional Services
Agencies and consultancies across from consulting to civil engineering have been among the most enthusiastic adopters of section-by-section proposal drafting. I have seen AI compress days of work into hours, but the most important part is that a human ensures the firm still sounds like itself. That’s why the best AI workflows in this category send a request for approval with the link to the drafted document straight to the reviewer, often simultaneously updating the task status within project management software.
The Common Architecture
Strip away the industry context and the workflows look remarkably similar. The agent does the reading, the searching, the summarizing and drafting. The human owns the moment of judgement, as every one of these settings is a high-trust or highly regulated environment.
From Automation to Intelligence: How Leading Teams Are Rethinking Operations
Register
Ai4 2026
Register
Content Leaders Collective: When Your Documentation Tools Can't Keep Up
Register
How to Win the War for Agentic Citations: The AEO Playbook You Need Now
Register
What is perhaps most interesting is how exactly enterprises are building the pause into the AI workflow. They don’t ask employees to fill out a separate form or log into another console. Instead, they route the human checkpoint into the tool the reviewer already uses: Slack, Teams, Outlook and so on. Human in the loop does not have to add friction and the examples above show that it can be seamless.
It’s no coincidence that the heaviest adopters are concentrated in law, financial services, investment management, healthcare and other highly regulated industries.
These are industries where the cost of a wrong answer is structurally higher than the cost of a slow one. Therefore, before deploying any AI system, CIOs and tech leaders in these regulated industries should ask themselves: In this workflow does human judgment create the most value and how do we make it as seamless to apply as possible?
Human in the loop is not a compromise with the vision of AI transformation. In regulated and high trust industries, it is the very vision that CIOs should pursue for safe, trustworthy AI systems that will provide long term value.
Editor's Note: The push for autonomous AI is running into a practical reality...
- Human-in-the-Loop Isn’t Optional in Agentic AI — AI agents can do more work, but humans still own the judgment. Here’s how to design HITL without creating new bottlenecks.
- Authorization Before Autonomy: A Governance Model for Agentic Commerce — As AI agents move from recommendations to execution, enterprises need controls for identity, permissions, context, auditing and rollback.
- Agentic AIOps: How to Build the Guardrails for Autonomous Infrastructure — AI agents can remediate outages in seconds. The real challenge is making sure they don’t turn small incidents into P0 disasters.
Learn how you can join our contributor community.