TL;DR AI models now help build the next AI models, and the cost and speed of marketing research change with them.
- Two labs report models that took part in their own development, MiniMax with M2.7 and OpenAI with GPT-5.3 Codex. Karpathy's open-sourced AutoResearch shows the same loop working on a single GPU.
- Research cycles are getting shorter. Monitoring and scanning work that took weeks now takes hours, and the models improve every few months.
- Autonomous experimentation lets a team complete ten experiments while a competitor completes one, and that learning speed is the durable advantage.
- Proprietary intelligence built on your own customer, category, and campaign data is the differentiator a late mover cannot buy.
- Build AI workflows now, invest in your own data, and treat AI as a system that learns.
Leopold Aschenbrenner, a former OpenAI researcher, plots AI capability over time in a graph. The curve is steady from 2018 to about 2024, then it steepens sharply.
The x-axis is time. The y-axis is effective compute, a proxy for how capable these systems are. The milestones are marked: GPT-2 wrote like a preschooler, GPT-3 like a schoolchild, GPT-4 like a smart high schooler. The next milestone on the chart is automated AI research. The confidence band after that point is very wide.
The evidence that we have entered the early phase of this curve arrived over the past few months.
The MiniMax release
In March 2026, the Shanghai AI lab MiniMax released a model called M2.7. The build process matters more than the release: the model participated in its own development. It identified its own performance gaps, planned changes, and ran improvement cycles inside an automated loop. MiniMax reports that the model handled 30 to 50 percent of the reinforcement learning team's daily research workflow, and that over 100 fully automated analyse-plan-modify cycles lifted internal evaluation scores by roughly 30 percent.
OpenAI has described GPT-5.3 Codex in similar terms: early checkpoints of the model were used to debug its own training and diagnose test results.
Andrej Karpathy, former Tesla AI director and OpenAI founding member, open-sourced AutoResearch: a tool where you hand an AI agent a training script and let it propose changes, run experiments, evaluate, and iterate on its own, overnight, on a single GPU, without a human in the loop.
Together they describe one pattern: AI is helping to build better AI.
Consequences for marketing and brand teams
Most coverage of recursive self-improvement, the pattern of models helping to build the next models, is written for AI researchers. The commercial implications concern marketing, brand strategy, and analytics teams, and they apply now.
1. Research cycles are getting shorter
A traditional segmentation study is expensive and slow. The same is true of demand-space mapping, brand trackers, and competitor analysis. Because this research is expensive and slow, it is infrequent, and the insight is often stale by the time it reaches a brief.
AI systems are improving at the rate described above, so the cost and time of each research cycle fall with every model release. Research is the "understand" stage of the marketing value chain; marketing before and after AI covers how AI changes each stage.
My own evidence is smaller than a segmentation study. I operate systems today that do in hours what used to take weeks: competitor monitoring, contact enrichment, market signal scanning across hundreds of sources at once. They are production workflows in daily use. The models powering them are materially better than the ones I used three months ago.
A planning cycle that assumes research is expensive and slow is working from an assumption that no longer holds.
2. Experimentation is getting faster
Good strategy often takes too long to reach the market.
Karpathy's autonomous experimentation transfers directly to marketing operations. The loop itself, in which a system designs an experiment, runs it, reads the results, adjusts, and runs again overnight without supervision, exists as working code for machine learning training. A marketing version would have to be built.
Apply that pattern to campaign testing, pricing, creative optimisation, and media mix. A team that runs 50 experiments while a competitor runs 5 learns faster, and while the models themselves improve every few months, learning speed is an advantage a competitor cannot copy quickly.
Teams that connect strategy to automated experiments, with results feeding back into strategy, will learn faster than teams that do not, and the difference grows each cycle.
3. Proprietary intelligence as a lasting advantage
Generic AI is a commodity: your competitor has access to the same ChatGPT, Claude, and Gemini, answering the same questions from the same training data.
A model tuned on your customer interactions, your category data, your campaign history, and your competitive signals is a different asset. It is proprietary.
Brands that start building proprietary intelligence now, even simple versions, gain an advantage a late mover cannot match by purchasing the same SaaS tool as everyone else. Your data is unique, and the intelligence built on it should be too.
Priorities for marketing and brand teams
Nobody knows exactly how fast this goes. The direction is the same in each of the three cases above.
The priorities for marketing and brand teams follow:
Build AI workflows now. The advantage grows the longer a team uses these workflows. A team that has run AI-assisted research, brief generation, and campaign analysis for 12 months operates at a different speed from a team starting next year, because the tools improve while you learn them. Our Marketing in the Age of AI series covers which workflows to build first.
Invest in your own data. Proprietary intelligence needs clean data and clear use cases more than it needs a data science department.
Treat AI as a system that learns. A model does not improve by itself between uses. It improves when each cycle's results are fed back into the next, so build that loop into the workflow for your customers, your category, and your competitors.
The pace of improvement is increasing, and the teams that adopt these systems early will be hard to catch.
We built NavigatorLab to answer one category question with AI-assisted research: where the growth is in your market.
Sources and further reading
- Leopold Aschenbrenner, "Situational Awareness: The Decade Ahead" (situational-awareness.ai)
- MiniMax, "M2.7: Early Echoes of Self-Evolution" (March 2026); coverage in VentureBeat
- Andrej Karpathy, AutoResearch (github.com/karpathy/autoresearch)
- OpenAI, GPT-5.3 Codex release notes