Coding Agents as First Control Point
Core Automation
This claim is really about where the first durable control point in work automation sits. If the winning agent stack is mostly software that reads codebases, writes code, runs tests, uses terminals, and improves itself through engineering loops, then Reflection starts closer to the money and the user workflow than Core. Core is trying to automate the lab that invents better systems. Reflection is trying to automate the work surface where those systems get used first.
-
Core is still a lab first company. Its own material says the near term product is automating internal research, and as of May 2026 it had no public API, pricing page, signup flow, or commercial product. That means its path to general work automation depends on first turning internal research tools into something buyers can deploy.
-
Reflection is built around the opposite wedge. Its company profile describes coding agents as the starting product, and its website frames autonomous coding as the root node for broader computer based work. That matters because coding already gives an agent a concrete environment, code, tests, terminals, repos, and measurable success or failure.
-
Sakana shows the middle path. It has already published AI Scientist work on end to end research automation, including a Nature paper and an earlier peer reviewed result, which validates the research automation thesis. But it is also attaching that research to products and partnerships, which is the step Core still has ahead of it if coding agents become the faster distribution channel.
The market is likely to move from coding into adjacent desk work that looks like coding under the hood, research, analysis, ops, and internal tooling. If that happens, the winners will be the companies that own the agent runtime where work is executed every day, not just the lab that discovers the next learning loop. That is why the coding wedge could let Reflection expand outward faster than Core moves downward from research automation.