WorkBuddy vs Codex: The Answer I Keep Typing Into My DMs

The question that keeps coming up in my DMs, over and over, is always the same: WorkBuddy versus Codex—which one is actually better?

Every time I see that question, my first instinct is to flip my phone face-down. It's not that I don't want to answer; it's that the question itself is framed wrong. Asking "which is stronger" off the bat is like asking whether a truck or a sports car is stronger. Tell me first: are you hauling twenty tons of building materials to a construction site, or are you lining up for a lap on the Nürburgring?

Tools don't have winners and losers; they have positioning. And in the Chinese market, the real dividing line has never been benchmarks—it's whether you can actually get the thing running.

I Use Both, but the Division of Labor Is Rigid

No beating around the bush—here's my actual daily workflow: complex engineering work, the kind where I need to stare at a terminal and debug line by line, goes to Codex. Making a PPT, scheduling meetings, remotely juggling a few documents—the "I don't want to open a terminal" stuff—WorkBuddy is the smoother fit.

The work from my August 1st post, "Shipping Three Platforms in Two Hours with Codex"—I tried handing it to WorkBuddy, and it couldn't even touch the problem. The reverse holds too: last time I asked an admin colleague who had never opened a terminal to install Codex, she stared at the black window for forty seconds, then sent me a sticker that said, "Did I just break my computer?" Forget it.

So my selection logic fits in one sentence: regular white-collar worker, go WorkBuddy; developer or someone who wants to dig deep into AI, go Codex. Different positioning—forcing a head-to-head is meaningless.

The Barrier to Entry Is the Real Wall at Home

Review articles love to compare model capabilities and benchmarks. I get it, but in the real-world scenario of actually getting people up to speed, all of that goes to the back of the line.

The single most important question is: can this person actually get the tool running?

I've taught far too many people to install Codex. The network and account hurdles alone will turn away 99% of people. Installation requires a terminal, registration requires an overseas phone number or email, payment requires a foreign-currency card, and you have to sort out the network yourself. Every single step is a wall. I've seen too many non-technical users in China die at the install-and-register stage, never even seeing what the product looks like.

WorkBuddy? From my hands-on testing, you open it and it just works. No "first register an overseas account" prerequisite.

So my advice right now is blunt: if you're teaching people in China to use AI, go with WorkBuddy, eyes closed. Legal, compliant, reasonable. No explanation needed.

This isn't to say Codex is bad. But when it comes to "getting ordinary people to actually use AI," whether you can use it matters far more than how powerful it is. A top-tier tool you can't get running has zero value. I've been in that trap, and I'm not exaggerating.

Signals in a Few Public Posts

This week I came across several public discussions about WorkBuddy and Tencent's enterprise AI strategy. There's a lot of signal in them. I wasn't at the event, and I'm not in a rush to buy enterprise seats, so let me lay out the verifiable claims first, then layer in my own read.

Positioned "After QQ and WeChat"

The version that's been circulating the hardest in the community: WorkBuddy is the third strategic-level product in Tencent's history, after QQ and WeChat, and it's going after the enterprise AI entry point.

I take that with a grain of salt. "Strategic-level" ultimately gets validated by the market; slogans don't count. But the direction is clear to me: if Tencent is truly betting at this scale, it's not a "let's dip a toe in" tool product—it's an entry-point play for the next decade.

The common playbook for entry-point products is to lock down organizational procurement, compliance, and collaboration first, with developer experience coming later. That's also why it feels more like an office product and didn't go down Codex's developer-native route—I'd guess that when Tencent is pitching this internally, the first slide probably says "procurement workflow," not "developer experience."

The Three-Stage Framework for Enterprise AI Transformation

The path that keeps showing up in public discussions, compressed, is basically three stages:

• The super individual awakens first: AI does the work, humans step back into decision-making roles, and productivity jumps tenfold.
• Capability overflows into the team: you replicate the method across the team, and overall efficiency gains 30% to 50%.
• Experience sinks into the organization: the entire organization gains another 30% to 50%. Get here, and you've won.

There's a big pitfall hiding here, and I've seen it more than once: the boss sees an employee's productivity jump tenfold, and the first reaction is often to cut headcount. That's the most expensive move you can make—it's literally pushing your most valuable people out the door.

Flip it around: proactively awakening and training yourself into a super individual is the real lever for breaking through. Whether you use WorkBuddy or Codex is a second-step question. Whether you've actually integrated AI into your real work is the first step.

A New Role: AIBP

We used to have HRBP. Next up is AIBP (AI Business Partner).

When people talked about FDE (forward-deployed engineer) before, the underlying logic was still outsourcing: send someone into the company, execute the transformation, and when they leave, the capability leaves with them.

AIBP is positioned as the organization's own person: the boss deploys them into a department to help the whole department go agentic; once productivity is up, they move to the next department. This kind of AI driver that grows from within the organization is extremely scarce right now. My read: AIBP is better than FDE because it doesn't depend on external blood transfusion.

There's also a pretty blunt supply-demand mismatch in the public discussion: big tech and the enterprise side have touched a huge volume of demand, but the supply of people who can actually use agents to solve specific scenarios can't keep up. This means:

• People who can actually use agents are worth more than people who can only talk about "empowerment."
• The ecosystem gap isn't in the slogans; it's in execution.
• If events like "Walk into Tencent," architect Q&A sessions, or on-site problem-solving with business leads start popping up, they're essentially plugging this supply chain.

The same logic applies to my own community: on one side, I help people go deeper with agents; on the other, I connect the people who use them well to real enterprise scenarios. When the deep users make money and the enterprises get results, the flywheel actually works.

A Blunt One: It Serves the Boss Who Pays

Signals covered. Time for the sharp part. Put its product form next to Codex's developer-native route, and there's a sentence that's not very polite, but I have to say it:

WorkBuddy is a product that satisfies the paying manager but hasn't yet shown enough respect for the people actually doing the work.

It hits every expectation that "pays but doesn't use" bosses have for AI software: the interface looks like a serious enterprise tool, data stays in-country, it can generate reports, it can map to the org chart, and the procurement process works. A boss watches one demo and is ready to sign. I've had more than one procurement lead tell me, "I don't use this thing, but I have to buy it."

But for the people actually doing the work day in, day out, the gap between it and Codex is a gap at the core and infrastructure level:

In yesterday's post, "Your Company Laptop Might Not Even Get Through the Agent's Front Door," I wrote: an agent without a runtime is just a chat box. WorkBuddy's current form is still a significant distance from the real value of AI-native—letting AI genuinely take over an entire stretch of work while humans step back into decision-making roles.

It meets the boss's expectations, but it hasn't yet delivered on AI-native value. I have to draw the line here, or all those earlier compliments about "convenience" and "compliance" turn into an advertorial.

The Window Is Real: Six Months to a Year

Criticism aside, my overall read is optimistic: WorkBuddy has enormous potential.

Three reasons:

First, it's an agent tool that fits the Chinese context better. Legal, compliant, out-of-the-box, data stays in-country—these three are all deal-breakers in domestic enterprise procurement, and it clears all of them. I've seen too many overseas tools die on the four characters "data leaving the country." WorkBuddy sidesteps that mine by default.

Second, the ecosystem is already partially connected across domestic and international lines. What comes next is a race for ecosystem positioning: whoever connects more wins, and whoever locks in user mindshare first takes the lead. The enterprise AI entry point is first-come, first-served.

Third, the window is still open. Yesterday I wrote that the Harness layer is about to become obsolete and that laptop-level agent tools will be replaced by the next generation of runtime very soon—but that's the timeline for frontier players. For the domestic market, the window is at least another six months to a year. In that time, whoever nails "low barrier + compliance + ecosystem" will have built up enough user assets before the next paradigm shift.

Five Sentences

• Don't ask which is stronger; ask what you're using it for. Complex engineering, pick Codex. Daily office work, pick WorkBuddy. Different positioning means no winner.
• The real dividing line at home is the barrier to entry. Network and account hurdles alone will block 99% of non-technical users from even getting through Codex's front door.
• In public discussion it's been slotted into a strategic position after QQ and WeChat. It's going after the enterprise AI entry point. Direction matters more than slogans.
• The blunt knife: it satisfies the paying boss, but it's still a distance from AI-native's real value. The gap is at the core and infrastructure level.
• The window is six months to a year. Nail low barrier + compliance + ecosystem, and you'll own mindshare before the paradigm shift.

In plain terms: teaching ordinary people to use AI, go WorkBuddy, eyes closed. Wanting to make yourself stronger, sit down and go deep with Codex.

Tool versus tool is a false question. Your own positioning is the real one: do you want to teach a hundred people to use AI, or do you want to make yourself a hundred times deeper?