MGS 2026 Recap: How ChatGPT Is Changing the Marketing Workflow
Your team has rolled out several AI tools, so why does the way you work look exactly like it did a year ago? If drafts got a little faster but research, reporting, and sharing still move from person to person by hand, the payoff from AI stays small.
The daglo marketing team attended Modern Growth Stack (MGS) 2026 in Seoul, a growth conference hosted by AB180 under the theme "Stack AI, Rewrite Everything." This post recaps "How ChatGPT Is Changing the Marketer's Workflow," a MAIN HALL session on July 21, 2026 by Dongjae Lee of OpenAI. We transcribed the talk with daglo, reread it, and pulled out what marketers can put to work right away.
Why the ChatGPT marketing workflow needs a redesign
Redesigning your ChatGPT marketing workflow means going beyond AI as a drafting assistant: the AI does the work first, and people step into the loop only when judgment is needed. Without that shift, the speaker said, even great AI only delivers local optimization.
The root problem is speed. New models used to arrive about every 15 months on average; now the cycle is much shorter. People and companies can't keep up with how fast models improve, and the market has split into three groups:
- People who know AI matters but are too busy to use it
- People who have started but haven't applied it to their core work
- People already using it in core work and seeing real gains
For marketers, the speaker said, AI is both a work partner and a new customer touchpoint, since more and more users now look for information through AI.
From chatbots to agents: three levels of agents
The speaker described 2022 to 2024 as the era of reactive chatbots that answer questions, and 2025 onward as the era of agents that keep context and act without being asked. He broke agents into three levels.
| Level | What it does | Example |
|---|---|---|
| GPTs | Purpose-built chat agents trained on company knowledge and templates | Drafting a launch plan from internal docs. Enterprise usage up about 19x year over year |
| Workspace Agent | Connects multiple apps on its own and finishes delegated work | Web research, analyzing cases in Google Drive, writing a Notion brief, scheduling, and posting to Slack in one run |
| ChatGPT Work and Codex | Takes on almost any task you do on a computer | Reads desktop files or code when needed and completes the job |
Workspace Agent is where AI starts acting like a teammate. Three traits make that possible:
- Built in plain language: No developer required. The best builders aren't a central team but the people who know the day-to-day workflow. Marketers build the best marketing agents.
- Learns fast: Show it a screenshot of the output you want and it figures out which tools to use, and that learning builds up across the whole organization.
- Runs in the cloud: It keeps working after you close your laptop or log out, effectively a coworker that works around the clock.
The speaker, who isn't a developer, built his own agents for planning the next day, summarizing overnight product updates, and preparing customer follow-ups, each in under 30 minutes.
Delegation isn't just for developers anymore
Inside OpenAI, 99.8% of output tokens now come from Codex and ChatGPT Work. In non-engineering teams such as finance, recruiting, and legal, Codex accounts for 60 to 90% of usage. Agents are less a developer tool than a way for everyone to delegate work.
Among enterprise customers, output through Codex has passed 50%. OpenAI's own marketing team uses agents from research and data visualization to campaign planning, A/B testing, and trend monitoring, and turns repeated workflows into skills. Each marketer gained nearly 10 times more time and test coverage, and the wait for the data team disappeared.
What matters more is using that time to run iterations far more often and improve results.
In a demo, ChatGPT Work built campaign performance slides from an overview in Drive and, given only a Slack channel name, shared them with leadership along with an executive summary. All of this depends on connection: file storage, messaging, calendar, email, and even internal legacy tools through custom MCP. The speaker also mentioned ChatGPT Ads, piloted in Korea since June.
Five principles for working well with agents
Beyond prompt writing, give an agent a clear goal, the context and constraints it needs, and an explicit "done when" condition. With the same agent and the same data, these choices make a visible difference in output quality.
- Give the agent a clear goal.
- Explain the context and constraints in detail.
- Define a "done when" condition so it knows when to stop.
- Start with one strong agent before adding sub-agents.
- For long, complex tasks, have it understand and plan before it acts.
If you're not sure what to delegate, ask the AI that, too. The speaker regularly asks, "Based on our conversations so far, diagnose which parts of my work are bottlenecks." Recent suggestions included a ready-made package of answers to common customer questions and modular meeting slides.
Takeaways: a checklist to try tomorrow
The core message: adopting AI and redesigning your workflow are two different things. Start by checking where your team stands today.
- Sort out whether you use AI only for drafts or actually hand it whole tasks.
- Pick one weekly task, such as a performance report, competitor monitoring, or meeting follow-ups, and delegate it to an agent.
- Whenever you delegate, write the goal, constraints, and "done when" condition together.
- Ask AI to diagnose where your bottlenecks are.
- Spend the time you save on running more experiments.
For recordings where it matters who said what, such as sessions with several speakers, daglo's Speaker ID is a useful feature. It remembers voices and fills in real names instead of "Speaker 1" from the next recording on, and daglo is the first service in Korea to offer it to every user. Learn more about Speaker ID
For the other sessions we attended, see our full MGS 2026 recap. Event details are on the official MGS 2026 site.
#MGS2026 #ModernGrowthStack #ChatGPT #MarketingWorkflow #AIAgents #Codex #GrowthMarketing #daglo
Where talk becomes work.
Get started with 1,000 credits a month — no credit card required.