Agents do the work. People sign off.
LHSG builds and runs systems where AI agents do the work and people approve it. With organizations that want to work the same way, we start by deciding what to hand over and how to check it.
01What to hand over
Start with work whose result can be checked.
We do not hand agents just anything. Work comes first when there is a way to tell whether the result is right. Here are three examples.
-
Development
What the agent does
Writes a plan, then builds in an environment separate from the original.
How it is checked
Checked with tests and on the real screen.
-
QA
What the agent does
Runs the tests and plays through the build.
How it is checked
Checked against a checklist.
-
Review
What the agent does
Reviews changes made by another agent.
How it is checked
The evidence for the review is recorded.
QWhat is AX?
AShort for AI transformation. It is more than handing out AI tools: what to delegate, how results are checked, and who is accountable are decided again for working with AI.
02How it works
A person signs off twice. Agents do the rest.
One task moves through six steps. A person signs off at two of them: before the work starts and before it is applied. Try approving the sample below.
Sample task
Add input validation to the contact form
- 01 Plan Agent Writes down what will change and how. Queued
- 02 Scope sign-off Person Reads the plan and approves the scope. Queued
- 03 Build Agent Works in an environment separate from the original. Queued
- 04 Verify Agent Checks with tests and on the real screen. Another agent reviews. Queued
- 05 Apply sign-off Person Looks at the result and the evidence, then approves applying it. Queued
- 06 Record Agent Records what was done and what could not be verified. Queued
The sample task is queued.
QWhat happens when an agent gets it wrong?
AThe flow assumes it can. Work happens in an environment separate from the original, results that fail verification are not applied, and a person signs off before anything is applied.
03What we built
We build it and use it before we recommend it.
Our AI transformation starts with what we have built ourselves: an agentic system for agents to work in, and the products we make with it.
Technology
Agentic System
A system that brings together the agents that take on work, the harness around them, and the points where a person signs off. The six steps above run inside it.
-
Goal Drift Control
Keeping long-running agents on the original goal
Keeps the original goal and completion criteria in a ledger while AI works, and checks tool calls and completion claims against approvals and evidence.
-
Recursive Harness Self-Improvement
RSI at the harness level
An experiment in which the agent proposes changes to its own harness, with selection and final evaluation kept separate, to see whether real development tasks improve.
-
Spec-Driven Agentic Development
Software development on an AI-native SDLC
Connects planning, design, and post-build review through one record. Requirements and scope are written down, and the screens that get built are checked and measured against them.
-
Agentic Game Development
Game development with agents in the loop
On every turn of game development, picks the checks that fit the request and the change, and passes the results to the next turn. Compares Unity lighting, inspects and drives the running UI, and captures the screen.
-
Agentic Marketing
Marketing run with agents
Plans targeting, positioning, channels, and campaigns, produces copy and content, and tracks posting schedules and results, together with agents.
Products
Made with the agentic system
-
SaaS
Web services
We build web services with agents, from planning to operations.
-
Games
AI-native games
AI is a core mechanic, not only a production aid. Characters decide and act on their own.
-
Content
Characters · Voice · Writing
We use AI for character art and motion, voice, and writing. We also build tools for writers to manage manuscripts and settings.
QWhat is an agentic system?
AA system built so that AI agents take on work and carry it through. It is made of agents and a harness, which defines what an agent knows, what it may do, the order work moves in, and how results are checked.
04Company
A company that has built software since 2008.
We started with cloud storage and a big data platform, then built set-top box and smart TV interfaces for broadcasting and telecom, mobile apps, and enterprise systems. Each time the technology changed, we learned it by building. Today we work with AI agents.
- Founded
- 2008
- What we do
- AI transformation (AX) · Agentic systems, SaaS, games, content
- History
- 2008Incorporated · Selected for a new-technology incubation program
- 2009Grand prize, lab start-up support program (SMBA Administrator's Award) · Certified venture company
- 2010Selected for an SME technology innovation start-up project
- Fields
- Smart STB
- Mobile
- Enterprise App
- Cloud
- XR
- Metaverse
- Worked with
- LG CNS
- C&M
- Gangwon Broadcasting
- Geumgang Broadcasting
- Hangang Broadcasting
- TV Storm
- RealNetworks
- LOEN
- Korea Asset Management Corporation
- SK Broadband
- Acryl
- Contact
- contact.lhsg@gmail.com
05Contact
Tell us by email what you would like to hand over.
If you include the three things below, we can start by working out the scope and how to check it.
- 1The work you do today, or the product you want to build
- 2The work you would like to hand to agents
- 3How you check results today
contact.lhsg@gmail.com
If your mail app does not open, copy the address and write to us.