Agentic coding unlocking ambition, but custom crypto is still terrifying
A post called 'Fast and Hard Code' argued that agentic coding tools are making previously infeasible projects tractable: building your own framework, database, or game engine no longer requires years of dedicated work. The top comments agreed on the ambition-unlocking effect but immediately flagged the obvious risk: agentic tools also make it easier to produce custom cryptography, which is one of the most reliably dangerous things a developer can do. That specific phrase, 'custom cryptography,' drew visceral reactions.
The pattern: the capability expansion from AI coding agents cuts both ways. Projects that used to require deep expertise are more accessible, but so are projects that should require deep expertise and don't get it. The guardrails are not keeping up with the capability.
There is a real debate in the comments between people who see this as liberation (finally can build the thing that was too big) and people who see it as a footgun factory (the things that were too hard were hard for good reasons).
So what?
If you are using agentic coding tools to tackle harder problems than you previously could, that is genuinely exciting. But the ceiling on what 'too complex to get right' means has not moved. Security-critical code, distributed systems correctness, and anything touching cryptography still require human expertise that the agent does not have. Use the tools to move faster on the parts that can be wrong and fixed; be more careful, not less, on the parts that cannot.