It is hard to communicate how much programming has changed due to AI in the last 2 months: not gradually and over time in the "progress as usual" way, but specifically this last December. There are a number of asterisks but imo coding agents basically didn’t work before December Show more
How software gets built next
AI agents are taking over more of the typing. The job isn’t going away: it’s moving up the stack, to deciding what to build, directing the agents that build it, and proving the result works.
HookDrill’s thesis, October 2026. A forecast, not a certainty.
It has already started
- Gartner expects 90% of enterprise software engineers to use AI code assistants by 2028, up from under 14% in early 2024.
- In Stack Overflow’s 2025 survey, 84% of developers use or plan to use AI tools, and 66% are frustrated by answers that are “almost right”.
- Anthropic’s 2026 report describes developers handing more implementation to agents while staying in charge of setup, supervision, review and validation.
Adoption and trust are separate problems. A plausible implementation is not a correct, secure, maintainable system, so as agents write more code, tests, review and clear requirements matter more, not less.
The workflow, step by step
One plausible way a feature gets built around 2031. Agents carry more of every step; a person stays accountable for each one.
Define the outcome
Start from the user, the goal and how success is measured, not a list of files to edit.
Ask clarifying questions and surface edge cases you skipped.
Write the specification
Review and correct user stories, data models, API contracts and acceptance tests.
Turn the goal into a structured, increasingly executable spec.
Design the system
Choose the architecture, the boundaries and which existing services to build on.
Propose options and wire up the integration scaffolding.
Delegate the build
Split the work, set the rules, review the risky changes and resolve conflicts.
Implement in parallel: frontend, backend, data, tests, docs.
Verify the behaviour
Judge the evidence: test results, traces, measured performance. Not a fluent “done”.
Run unit, end-to-end, accessibility and security checks; analyse failures.
Deploy with guardrails
Set the boundaries: credentials, environments, spending limits, who approves what.
Prepare releases, deploy to staging, promote when the policy allows.
Monitor and maintain
Own the incidents, the trade-offs and the access an agent is allowed to have.
Spot anomalies, summarise incidents, propose patches as pull requests.
What changes for the engineer
| Less of | More of |
|---|---|
| Implementing routine changes by hand | Defining tasks and delegating implementation |
| Writing boilerplate and repetitive tests | Reviewing coverage and validating behaviour |
| Knowing every framework detail from memory | Understanding concepts and judging generated code |
| Sequential handoffs between people | Coordinating parallel agents through shared interfaces |
| Reviewing every line the same way | Reviewing by risk, uncertainty and impact |
| Deploying through manual procedures | Designing safe, observable release automation |
| Delivering the requested feature | Owning outcomes, reliability, security and value |
None of this makes code literacy optional. Reading, debugging and reasoning about code is exactly what lets an engineer evaluate what an agent produced, and catch the error a polished explanation hides.
What builders are saying
People building with AI every day largely agree: the work is moving from typing code to deciding what to build, directing agents and checking what they produce. The craft doesn’t disappear; it moves up the stack.
A few random notes from claude coding quite a bit last few weeks. Coding workflow. Given the latest lift in LLM coding capability, like many others I rapidly went from about 80% manual+autocomplete coding and 20% agents in November to 80% agent coding and 20% edits+touchups in Show more
Prediction: The next 12-24 months, "UX-pilled" builders will be in massive demand. Who can create intuitive interfaces, web+mobile+desktop apps that "feel good," natural, fast, and far better than the competition. THIS will be the difference vs those building "just" with AI.
I think this group of reactions are still fundamentally rooted in a scarcity mindset of software. 2 years ago AI was botching autocomplete, today it is almost one shotting browsers and C compilers. Where is it in 2 more? 10? 20? Software so insanely cheap and abundant that Show more
AI will soon graduate from being a vibe coder to a software system creator Powerful AI agents will be able to design, develop, test, monitor and scale software systems These agents will build new features, fix bugs and handle tech support Coming in the next few months…
Most confusion about the future of software design stems from a confusion in terminology. My view: production design will increasingly be automated. The economic logic is self-evident — training machines to mimic and refine existing production practices is cheaper, faster, and Show more
@rauchg
Whether design belongs in Figma or Claude Design is a distraction from a bigger shift. 1️⃣ Design will become autonomous. More helpful to think of it as 𝙳𝙴𝚂𝙸𝙶𝙽.𝚖𝚍, used by your coding agents running your software factory. 2️⃣ Specialized “personal” design tools generated
Posts are shown as published on X. Quoted for context, not endorsed.
What stays hard
AI will get better at all of these. They stay risky to hand over blindly because they involve uncertainty, consequences and reasoning about the whole system.
- Turning vague business goals into correct requirements
- Complex state, concurrency and distributed failure
- Protecting sensitive data and security boundaries
- Proving correctness where tests can’t cover everything
- Keeping large, old codebases consistent
Why HookDrill exists
The routine tasks juniors used to learn on are the first ones agents take. Skills that nobody practises fade quietly, and the people who will direct agents tomorrow still need to recognise good code when they see it.
HookDrill is a learning platform for the developers who will build software this way, starting with web apps. It trains both halves of the job.
Open nowDrills
Build real React components from a spec, by hand, with hidden tests as the judge. The fundamentals you need to read and repair what agents write.
Start practicingSoonBuilder Lab
Practise the other half: write the spec, design the system, direct an agent, and verify what it shipped. Same sandbox, same hidden tests.
See Builder LabSources
- Gartner: Top Strategic Trends in Software Engineering for 2025 and Beyond (July 2025)
- Anthropic: 2026 Agentic Coding Trends Report
- Stack Overflow: 2025 Developer Survey
The 2031 workflow is a reasoned extrapolation from current trends, not a prediction that every team will work this way. Survey figures depend on who responded; forecasts come from the organisations named.