Blog
Back to all posts
OpinionOutlookAI

Software Projects in the AI Age (2027–2028)

By late 2027, software development teams will look fundamentally different. Code is written by AI. Humans architect, review, and orchestrate. The question isn't whether this happens — it's how teams restructure around it.

This is my prediction for how software projects will be staffed and run in the near future. Not hype — a practical, realistic assessment based on where AI capabilities are heading.

The Three Project Tiers

Not all projects are equal. The team structure depends entirely on complexity. Click each tier to see how it works:

Small Projects

~1 week traditional effort

Team Composition

  • 1 Project Manager
  • No engineers needed

Workflow

  1. 1. PM takes requirements from stakeholders
  2. 2. PM translates requirements into prompts
  3. 3. AI agent builds the solution end-to-end
  4. 4. AI system evaluates if the problem is truly simple
  5. 5. Back-and-forth Q&A between PM and AI for clarification
  6. 6. PM reviews output and delivers to stakeholders
💡

"Fail-proof" — simple enough that AI handles it reliably. No deep technical understanding needed from the PM.

Onboarding Junior Engineers: The Hard Part

If AI writes all the code, how do you train the next generation of senior engineers? This is the critical unsolved problem.

My answer: use small projects as training ground. Instead of throwing juniors into large projects where they'd be overwhelmed, a manager pre-scopes requirements and the junior engineer implements with AI assistance.

  • It's slower for the team — but necessary
  • Without training juniors, there won't be future senior engineers
  • Think of it as an apprenticeship model
  • Juniors learn architecture, review, and orchestration — not manual coding

Skip this step, and the industry creates a talent cliff. You can't hire senior engineers if nobody ever gets trained up.

Skills: Past vs AI Age

The required skillset shifts dramatically. Select a role to see what changes:

Software EngineerSkill Transition

REMOVEDDeep coding skills (writing code manually)
REMOVEDLanguage/framework expertise (as primary skill)
STAYSDebugging
STAYSVersion control
STAYSTesting
STAYSCode review & reading comprehension
EVOLVESBroad technological overview→ Must understand the full stack
EVOLVESBasic architecture understanding→ Architecture expertise (CRITICAL)
NEWAI agent systems & agentic workflows
NEWPrompt engineering / AI orchestration
NEWSystem design & infrastructure (elevated)
NEWTechnology landscape awareness
NEWCommunication & stakeholder management
No longer needed Carries over Evolves New skill needed

The Transition Timeline

15%

Now (2026)

AI writes code, but humans still do most of the integration, debugging, and architecture. AI is a tool, not a team member.

55%

Late 2027

Small projects fully AI-driven. Medium projects need one senior engineer as conductor. The restructuring begins.

90%

2028+

Three-tier model is standard. Junior onboarding becomes the critical challenge. Teams are smaller but more senior.

% of code written by AI agents (estimated)

What This Means

Teams get smaller but more senior. The role of "software engineer" shifts from someone who writes code to someone who understands systems deeply enough to direct AI that writes code. Project managers on small projects become direct AI operators.

The junior engineer problem is real — but it's conditional. If AI agents keep improving at their current pace, they may eventually replace the need for junior roles entirely. In that scenario, the apprenticeship pipeline becomes irrelevant because the profession itself transforms beyond recognition.

But if agents plateau — capable enough to write code but not enough to fully own projects — then the industry must invest in junior training or face a talent crisis within a decade. The companies that figure out the apprenticeship model first will have a lasting advantage. Everyone else will be fighting over a shrinking pool of senior talent.


This is an opinion piece reflecting my current thinking. The timeline could shift. The specific team structures will vary. But the direction — AI writing code, humans architecting and reviewing — feels inevitable. The only question is how fast.