After 40 years of computers helping automate factory lines, we're finally seeing AI tackle knowledge work in meaningful ways. But there's a catch - while language models can now handle complex tasks, most tools still fall short of truly automating our daily work. That's where Hunch, and their new pre-release product Overclock, come in with a fresh take on AI automation.

VMblog had a front row seat with Hunch while at the 62nd edition of the IT Press Tour taking place this week in San Francisco. 

The Team Behind the Technology

David Wilson, CEO and co-founder of Hunch, brings serious enterprise tech credentials to the table. His previous company, Cape Networks (acquired by HPE/Aruba), made network monitoring accessible for IT teams. Now he's targeting a bigger challenge - using AI to eliminate what he calls the "busy work tax" that plagues knowledge workers.

"If we think about the average knowledge worker's job, there's a lot of busy work that we could, in principle, automate away," Wilson explained during the IT Press Tour briefing. "So much of knowledge work is actually just content transformation from one system to another, with very little value add in that process."

Why Current Solutions Fall Short

Wilson outlined three main approaches people try today, and why they don't quite solve the problem:

1. Traditional automation tools like Zapier and IFTTT - They're very deterministic, and cumbersome to create flows that are often brittle and hard to maintain, explained Wilson.

2. ChatGPT hacks - "People now have their little pet prompts or GPTs they've created, and they'll go back to ChatGPT or Claude and use those to create micro-tasks. Then they're basically copying and pasting for hours."

3. AI-enhanced automation tools - While getting closer, these still don't let average users truly 'think about and describe that piece of work, and then not have to worry about it again.'

Enter Overclock: Plain English Automation

Overclock takes a different approach. Instead of complex flowcharts or rigid automation rules, it lets users describe tasks in plain English through what Wilson calls "playbooks."

"You can describe it in plain English, in a playbook that says, for example, 'at 8am go to these competitors, check prices' and have it as a plain English source of truth - almost like source code for that particular piece of work, but in English that is very easy to understand, easy to generate, easy to edit," Wilson explained.

How It Actually Works

The system uses a sophisticated multi-agent approach that starts with a simple user intent and transforms it into automated action. Wilson explained the process through an example:

"You start off with an intent - like 'monitor my competitors' - and it then writes a playbook that is step-by-step instructions for an agent to follow in English. It executes that playbook, tests it, and allows you to deploy it."

The system architecture includes several key components:

1.  Intelligent Agent Execution - Instead of having to articulate these automations in very complex flows or code, you're able to describe it in plain English. The system's agents can "fill in the gaps where need be, recover from any kind of failures along the way."

2.  Adaptive Learning System - The platform implements two distinct types of learning:

• Knowledge about the organization and its specific needs
• "Know-how" - learning how to execute tasks effectively "If it's able to navigate itself, it should be able to learn for next time," Wilson noted. "Not only it, but the whole system should be able to do that for your organization."

3.  Workplace Integration - "It needs to live where you work," Wilson emphasized. "For a lot of people, this basically means Teams or Slack and maybe as a fallback, email, WhatsApp, Telegram. It needs to come to you if it's having a problem, if you need to be able to provide that feedback.

4.  Multi-Agent Oversight - The system employs what Wilson calls a "watchdog" agent: "We actually have a separate agent that is sort of like a watchdog, looking at this thing as it's executing, to make sure that it is following the instructions and that it is doing the right thing."

5.  Knowledge Sharing Architecture - "Because a lot of this is now able to be expressed in plain English, it is much easier to share and to customize these automations than it has been before with these very flow-based solutions," Wilson explained. This allows successful automations to be easily adapted and reused across teams.

The system is designed to handle failures gracefully. For example, if a competitor's URL changes or a site is temporarily inaccessible, the agent can adapt and find alternative approaches rather than simply failing like traditional automation tools would.

This architecture represents a significant shift from traditional automation approaches, moving from rigid, predefined workflows to an adaptive system that can handle the complexities and variations of real-world tasks while continuously improving its performance.

Real World Applications

While Overclock is still in pre-release, the underlying Hunch platform already powers some impressive use cases. Wilson shared an example of a major advertising agency using it to analyze multinational brands:

"They use Hunch with a very sophisticated process... running thousands of agents to scrape every single URL on their sub-brands' websites in every country, then put that together into a database of all their products and services."

The system then finds commonalities, tensions, and messaging opportunities - work that "would be months and months for Accenture to do. And they're able to do it in a morning."

The Technology Stack

Hunch integrates with leading AI models and services, including:

• Multiple large language models
• Transcription services
• Image generation models
• Custom enterprise integrations

The platform uses a credit-based system that Wilson says keeps costs reasonable: "Most people don't pay a lot every month. It scales with value... if it's something that is really worth doing, like sharing out tons of posts and going through lots of different stages to get the best outcome, then it's going to be worth it."

Looking Ahead

Wilson sees a future where AI automation becomes as fundamental as infrastructure-as-code tools like Terraform: "It's almost like your knowledge work enterprise code that you can then use to create a lot of your routine tasks over time using AI."

The company is taking a bottom-up, product-led growth approach, letting the technology spread naturally as users share successful automations with colleagues.

The Bottom Line

For IT teams and enterprise users, Overclock represents an interesting new direction in AI automation. Rather than replacing knowledge workers, it aims to eliminate the "busy work tax" that keeps them from focusing on higher-value tasks.

The pre-release version is available now, though Wilson notes it comes with an appropriate warning: "If you don't like bugs and chaos, you should turn back." But for organizations looking to get ahead of the AI automation curve, it might be worth exploring despite any early rough edges.

The combination of plain English instructions, multi-agent oversight, and workplace tool integration suggests we're getting closer to AI automation that actually works the way humans do. And that could make all the difference in moving from interesting demos to practical enterprise solutions.