Maybe software stops being the default

4 hours ago

The author observes a progression from writing software, to AI writing software, to no software being written at all for certain tasks.
AI agents can follow instructions and perform tasks without creating durable software, replacing the need to design, deploy, and maintain persistent applications.
Examples include accounting reconciliation and timesheet management, where agents directly handle workflows using API keys and instructions instead of custom software.
Agents function like human delegates: given context, instructions, and access, they perform tasks without exhaustive formalization into code.
The decision to use an agent versus formal software depends on factors like task frequency, cost of mistakes, and ability to check output.
Software will remain for tasks requiring repeatability, determinism, and performance, but it may no longer be the default automation mechanism.
Many ad-hoc processes, internal tools, converters, and bespoke UIs may simply not be built as agents become more reliable.

AI's PC era is arriving, focusing on personal AI that is easy, free, and dependable, operating locally on devices.
Personal AI should not be a cloud service but a layer between users and destinations, using user data and models.
Current software clutters screens to capture attention, with apps prioritizing platform goals over user intent.

The author now relies heavily on AI agents for producing entire PRs in familiar areas, starting every change with an agent request and pushing the PR after a single editing pass.
Agents are used extensively for bug investigation, successfully diagnosing 80% of issues autonomously, though human expertise remains crucial for narrowing down complex bugs.
The author still writes PR descriptions by hand to signal human review and avoids using AI for Slack messages, ADRs, or blog posts, but uses AI for feedback on drafts.

The author initially doubted AI's coding abilities, treating it like Stack Overflow and manually reviewing code to maintain personal style.
Claude Code demonstrated AI's ability to handle maintenance tasks and reduce the need to learn specific frameworks like Laravel.
Newer AI models (e.g., Fable, GPT 5.6-Sol) now outperform the author in 80% of coding tasks, leading to acceptance and excitement about productivity gains.

AI agents cannot truly program; they are statistical models that produce increasingly hard-to-detect broken code.
The author experimented with agents for 6 months but found manual work more efficient; agents frontload progress but fail at polish.
AI is useful for prototypes and better than Google for searches but is not a substitute for software engineers.

Adopting a new tool involves three phases: inefficiency, adequacy, and life-altering discovery.
Chatbots are limited for coding; use agents that can read files, execute programs, and make HTTP requests.
Reproduce manual work with agents to build expertise, breaking tasks into clear, actionable steps.

AI's PC era is arriving, focusing on personal AI that is easy, free, and dependable, operating locally on devices.
Personal AI should not be a cloud service but a layer between users and destinations, using user data and models.
Current software clutters screens to capture attention, with apps prioritizing platform goals over user intent.

The author now relies heavily on AI agents for producing entire PRs in familiar areas, starting every change with an agent request and pushing the PR after a single editing pass.
Agents are used extensively for bug investigation, successfully diagnosing 80% of issues autonomously, though human expertise remains crucial for narrowing down complex bugs.
The author still writes PR descriptions by hand to signal human review and avoids using AI for Slack messages, ADRs, or blog posts, but uses AI for feedback on drafts.

The author initially doubted AI's coding abilities, treating it like Stack Overflow and manually reviewing code to maintain personal style.
Claude Code demonstrated AI's ability to handle maintenance tasks and reduce the need to learn specific frameworks like Laravel.
Newer AI models (e.g., Fable, GPT 5.6-Sol) now outperform the author in 80% of coding tasks, leading to acceptance and excitement about productivity gains.

AI agents cannot truly program; they are statistical models that produce increasingly hard-to-detect broken code.
The author experimented with agents for 6 months but found manual work more efficient; agents frontload progress but fail at polish.
AI is useful for prototypes and better than Google for searches but is not a substitute for software engineers.

Adopting a new tool involves three phases: inefficiency, adequacy, and life-altering discovery.
Chatbots are limited for coding; use agents that can read files, execute programs, and make HTTP requests.
Reproduce manual work with agents to build expertise, breaking tasks into clear, actionable steps.

Hasty Briefsbeta