Jediah Katz
A former Cursor engineer focused on Context Engineering and Agent Harness research within AI engineering paradigms, who has given a presentation at AI Engineer NYC.
Timeline (last 90 days)
Katz points out that there are three things in the harness that models cannot absorb: infrastructure and routing, trust and safety, and efficiency (cost optimization).
Katz predicts that within 6 months, 90% of Agent work will run outside laptops (i.e., in the cloud).
Jediah Katz defines Agent Harness as the system layer that "wraps the model, steers the model, and connects it to the world," divided into the Runtime layer and the Context orchestration layer.
The "data flywheel" mechanism proposed by Jediah Katz: the harness makes harder tasks accessible → more users generate coding data → the data flows back to model labs → the model natively learns the skill → the harness removes the patch.
Jediah Katz cites "Hiram's Law": anything exposed to users, users will depend on, so the visibility of harness internal details to users should be reduced.
Jediah predicts that within six months 90% of agent work will run outside the laptop.
Jediah proposed the concept of a "data flywheel": the harness makes increasingly difficult coding tasks accessible, more users use agents to attempt these tasks, more coding data is shared with the labs, and once models natively learn these skills, the harness can delete what was originally added to make the tasks work.
Jediah Katz and his colleague Stefan wrote about Cursor's process of removing a large number of "guardrails," such as removing the logic that once required a small model to specifically apply code edits to files, as well as various coding-specific instructions.
Jediah Katz is a former Cursor engineer who spoke at AI Engineer NYC.
All Facts (9)
Katz points out that there are three things in the harness that models cannot absorb: infrastructure and routing, trust and safety, and efficiency (cost optimization).
50%UnverifiedJediah Katz defines Agent Harness as the system layer that "wraps the model, steers the model, and connects it to the world," divided into the Runtime layer and the Context orchestration layer.
50%UnverifiedThe "data flywheel" mechanism proposed by Jediah Katz: the harness makes harder tasks accessible → more users generate coding data → the data flows back to model labs → the model natively learns the skill → the harness removes the patch.
50%UnverifiedJediah Katz cites "Hiram's Law": anything exposed to users, users will depend on, so the visibility of harness internal details to users should be reduced.
50%UnverifiedJediah Katz is a former Cursor engineer who spoke at AI Engineer NYC.
50%UnverifiedJediah Katz and his colleague Stefan wrote about Cursor's process of removing a large number of "guardrails," such as removing the logic that once required a small model to specifically apply code edits to files, as well as various coding-specific instructions.
50%UnverifiedJediah proposed the concept of a "data flywheel": the harness makes increasingly difficult coding tasks accessible, more users use agents to attempt these tasks, more coding data is shared with the labs, and once models natively learn these skills, the harness can delete what was originally added to make the tasks work.
50%UnverifiedKatz predicts that within 6 months, 90% of Agent work will run outside laptops (i.e., in the cloud).
50%UnverifiedJediah predicts that within six months 90% of agent work will run outside the laptop.
50%