Codex-Maxxing: Beyond Code to a Persistent AI Work Partner

In the rapidly evolving landscape of AI coding assistants, one practitioner has pushed the boundaries of what's possible with OpenAI's Codex. Jason Liu, a developer and productivity enthusiast, has developed an approach he calls 'codex-maxxing'—transforming these tools from simple code generators into persistent work partners that can maintain context across days, weeks, and even months.

"I was already using coding agents a lot before Codex. Mostly, though, I used them through interfaces built for coding work: making diffs, changing repos, and shipping code," Liu explains. "Around November, I started pushing them into knowledge work too."

This shift represents a significant evolution in how developers interact with AI assistants. While most users treat these tools as conversational code generators, Liu has developed a system that treats them as persistent workers with memory, agency, and the ability to maintain long-running projects.

The Operating Loop: Beyond One-Shot Interactions

What changed Liu's behavior was learning to give work an "operating loop"—a durable thread, shared memory, tools that can act on his computer, ways to steer and resume tasks, and a surface where he can review artifacts directly.

"The more interesting shift is that it gives my work somewhere to live," Liu notes. "Codex is still excellent for coding, but the more interesting shift is that it gives my work somewhere to live."

This operating loop transforms AI assistants from reactive tools to proactive partners that can maintain momentum across work sessions.

Durable Threads: Long-Running Conversations

The foundation of Liu's approach is "durable threads"—long-running conversations that accumulate context, preferences, and decisions over time. These threads are periodically compacted to maintain efficiency while preserving essential context.

"I now keep a pinned thread for every important workstream I care about: my Chief of Staff thread, the Agents SDK, the OpenAI CLI, Codex for open source, one just to monitor Twitter," Liu shares. "These are not short chats. They are megathreads that I have been compacting for months. They accumulate history, preferences, and old decisions that I do not want to recreate every time I come back."

This approach represents a fundamental shift from the typical pattern of starting fresh conversations with AI assistants. By maintaining persistent threads, Liu creates continuity that allows his AI partners to understand long-term projects and personal preferences.

Voice Input: Capturing Unedited Thinking

Liu emphasizes that voice input isn't about speed—it's about capturing the unedited version of his thinking.

"The benefit is not speed. It is that the agent gets the unedited version of my thinking," he explains. "If I am planning a piece of work, I might say, 'I think there is some guy named Ben in Slack who mentioned this, I do not remember exactly what, just go look.' That sentence is too vague and annoying to type, but completely natural to say."

This approach allows the AI to work with the messy, incomplete version of human thought rather than just the polished, finalized versions we typically share.

Steering: Real-Time Direction

"Steering" is another key component—adding direction while the AI is already working rather than waiting for each step to finish.

"If I am reviewing a website, I can keep talking while I look at it: make this smaller, this copy is wrong, the spacing between these two things feels off, once this is done, open a PR, wait for the preview deploy, send the preview link to the person who needs to review it on Slack," Liu describes. "I do not need to wait for each step to finish before deciding the next one. I can keep adding intent while the agent is still working, then walk away with the queue already shaped."

This transforms the interaction pattern from "one prompt, one answer" to continuous, fluid collaboration where the human and AI work simultaneously on different aspects of a task.

Memory Systems: Persistent Context

Once threads became long-running, Liu needed a way to preserve memory outside any single conversation. His solution was an Obsidian vault that serves as persistent memory.

"The point of the memory system is to turn what the thread learns into an artifact I can inspect, edit, diff, and reuse," Liu explains. "Most of my long-running threads start in an Obsidian vault: vault/ ├── TODO.md ├── people/ ├── projects/ ├── agent/ └── notes/"

This approach creates a shared memory space that different threads can access, allowing knowledge to accumulate across conversations rather than being lost when threads end.

Computer and Browser Integration

Liu distinguishes between different ways the AI can interact with his systems:

• $browser for local web surfaces he wants to inspect and annotate
• @chrome for signed-in browser state and multiple tabs
• @computer for work that only exists as a GUI

"On my work machine, Twitter is logged into Safari. If I have @computer read Twitter there, I lose Safari while it works. @chrome is better when I want the agent to use several authenticated tabs in parallel without taking over the whole app I am using," Liu notes.

This granular control allows the AI to interact with different parts of his digital environment appropriately.

Heartbeats: Automated Task Management

"Heartbeats" are Liu's term for recurring checks a thread can schedule for itself, creating automated task management.

"A Heartbeat is a thread-local automation. You can say, 'keep an eye on this every few hours,' and the thread can schedule itself," Liu explains. "A thread can have multiple schedules, run until some condition is met, and adjust its cadence over time."

His "Chief of Staff" thread runs every 30 minutes, checking Slack and Gmail for unanswered messages that need attention. This creates a proactive system that can prioritize and prepare responses before he even checks his messages.

Goals: Long-Running Tasks with Verification

Liu distinguishes between weak goals ("implement the plan in this Markdown file") and strong goals with real verification criteria.

"Last week I tried to migrate the Python Rich library into Rust. Because the original project already had a large unit test suite, I could set a goal like: migrate Rich into Rust, but it must pass all the unit tests from the original library," Liu shares. "That test suite gave the run a real oracle: the Rust port was not done until it passed the same tests as the Python library."

This approach transforms AI-assisted work from vague assistance to verifiable progress toward specific, measurable outcomes.

The Side Panel: Where Work Happens

"The part of Codex I am most excited about is the side panel. It is easy to think of this as a place where previews happen. That undersells it. The side panel is where Codex stops being only a chat app and starts becoming the place the work happens," Liu emphasizes.

The side panel serves three key functions: inspecting artifacts, operating web surfaces, and reviewing changes. This creates a workspace where the human and AI can collaborate on the same object simultaneously.

Implications for the Future of Work

Liu's approach represents a significant evolution in how humans interact with AI assistants. Rather than treating these tools as conversational interfaces or simple code generators, he's developed a system that treats them as persistent partners with memory, agency, and the ability to maintain long-running projects.

"The more Codex gets places to remember, revisit, inspect, and act, the less my work dies between prompts. That is the change I care about. Not that an agent can write code for me, but that more of my work can keep moving after I leave," Liu concludes.

This approach suggests a future where AI assistants become true work partners that can maintain momentum across sessions, remember context over time, and collaborate with humans on complex, long-running projects. As these capabilities continue to evolve, we may see fundamental changes in how we structure work and productivity in the age of AI.

For those interested in exploring these approaches, Liu's detailed breakdown provides a roadmap for transforming AI assistants from simple tools into persistent work partners. The key insight is that the most powerful applications may not be in what these tools can do in a single conversation, but in how they can maintain context and continuity across the complex, multi-faceted work that humans actually do.