Future Tech/workflow
Most Production Code Will Be Written in English
I am building production systems with Claude Code writing most of the code. This is not a future prediction. It is the current state, scaling fast.
// By 2028 · high confidence · disruption 9/10
Prediction
// 2028
By 2028, the majority of new production code at venture-backed companies will be specified in English and generated by AI assistants.
What dies
- → traditional seo
- → manual soc triage
Who wins
- → Claude Code
- → Cursor
- → GitHub Copilot
The hook
A developer at a Series B startup told me they shipped their last quarter's features with two engineers and Claude Code. The same scope would have taken eight engineers in 2022. The leverage is not 2x. It is 4x.
Thesis. The bottleneck in software engineering is not typing speed. It never was. The bottleneck is judgment, architecture, and debugging. AI changes the production function for code, not the skill of being a developer.
The story
The setup
2021 to 2023. GitHub Copilot launches. Autocomplete on steroids. Useful but not transformative. The interaction model was still 'developer types, AI suggests, developer accepts.'
The chat era
2023 to 2024. ChatGPT and Claude can write functions. Developers paste code in and out of chat windows. Still tool-mediated, still copy-paste. Productive but not yet integrated.
The agent shift
2024 to 2025. Cursor, Windsurf, Claude Code. AI runs in the IDE with full repo context. Multi-file edits become routine. The unit of work shifts from a function to a task.
The autonomous PR
2025 to 2026. Claude Code ships unattended PRs. Cursor agents complete tickets. Junior dev tasks become AI tasks. The economics of staffing engineering teams change quietly but materially.
The new default
By 2028, 'write the code yourself' is the exception, not the rule, at most venture-backed companies. The developer's job shifts toward specification, review, and architecture. Code is the output, English is the input.
First signals (verify today)
Cursor at $100M+ ARR within 18 months. GitHub Copilot at 70%+ enterprise developer adoption. Claude Code shipping autonomous PRs.
Key data points
- GitHub Copilot launch: October 2021
- Cursor founding: 2022 (Anysphere)
- Claude Code launch: 2024 (Anthropic)
- Cursor ARR: $100M+ within 18 months
- GitHub Copilot enterprise adoption: 70%+
Contrarian angle
Most developer roadmaps assume AI coding tools augment senior engineers. The actual pattern is that senior engineers become more productive while junior engineering roles compress. The job pyramid flattens.
The flip side
What this kills
The paired obituary in Tech Graveyard.
Read the obituaryFAQ
Do junior developers still have careers in 2028?
Yes, but the on-ramp shifts. Junior tasks that used to be 'implement this function' are increasingly handled by AI. The new entry-level work is closer to 'review AI-generated code, understand and own the system.' Mentorship and ramp times change with it.
Is 'vibe coding' the same as production engineering?
No. Vibe coding (no specifications, no review, ship-and-pray) works for personal projects and prototypes. Production engineering still requires judgment about correctness, security, performance, and maintainability. AI accelerates the work; it does not eliminate the responsibility.
What skills become more valuable, not less?
Specification clarity, code review at speed, system thinking, debugging across abstraction layers, taste in architecture choices. These were always valuable; they are now the rate-limiting skills.
More from guptadeepak.com
Want the technical deep-dive behind this prediction?
Read the companion articleRelated predictions
More from the workflow desk.
// By 2027
high confidenceServer-Side Attribution Becomes the Only Attribution
The death of third-party cookies forces a complete rebuild of digital marketing attribution. Server-side wins. Most advertisers are not ready.
First signals: Meta CAPI at 70% advertiser adoption. Google Enhanced Conversions mainstream. CDP market growing 20%+ annually.
workflow · Disruption 6/10
// By 2029
medium confidencePersonal AI Clones Take Your Calls and Meetings
An AI agent trained on my emails, my Slack history, and my voice can already attend a meeting and brief me after. By 2029 every knowledge worker will have one.
First signals: Voice cloning at ElevenLabs and Cartesia indistinguishable from real. Meeting recap agents (Otter, Fathom) handling half of meetings. Claude and GPT-4o multimodal agents.
workflow · Disruption 9/10
// By 2030
high confidenceAI Becomes First-Line Medical Diagnosis
AI already outperforms most radiologists on specific tasks. By 2030, AI is the first reader, not the second. The clinical workflow rebuilds around it.
First signals: FDA approved 950+ AI medical devices by 2024. Aidoc, Viz.ai in production at major hospital systems. Google Med-PaLM 2 outperforming USMLE.
workflow · Disruption 9/10