Dis week, one Chinese AI lab publish one research prototype wey dey watch developers group chats and GitHub repos by imself, den e dey propose code corrections even wen nobody ask am. DeepSeek im Super-Agent talk open say im work never finish, but one line inside im document important pass di tool itself: na coding agents do most of di engineering. Di people wey dey build di replacement dey use di replacement to build am.
About di same time, one joke video wey dem call "Software Engineers 2028" begin waka among developers, imagining say di profession don reduce to watching machines and calming stakeholders. E touch people because e touch real worry. If agents fit read di chat, write di patch and open di pull request, wetin exactly remain for di job?
History get answer wey correct well well, and e start with di floppy disk. Di floppy disk no disappear because people stop to store data. E disappear because di function move go better media, and di object wey bin define di function don turn museum piece. Job titles dey follow di same physics. Di typist no disappear wen writing finish, e disappear wen writing separate from one dedicated machine and one dedicated person. Functions dey migrate; titles dey wait small den dem dey retire quietly.
Software engineering dey inside dis migration now. Di part of di job wey be pure transcription, turning clear intention into syntax, dey move go machines with remarkable speed. GitHub own research find say developers dey finish routine tasks dramatically faster with AI assistance, and every big lab now dey ship agent wey fit hold coding conversation. Di typing layer dey waka di floppy disk road.
But profession no be one layer. Remove typing and three harder layers remain. Di first na to decide wetin suppose exist: frame di problem, choose di trade-offs, know which requirement real and which one na stakeholder guess. Di second na verification: read machine-written code with di suspicion wey e deserve, because agent wey dey write fluently dey fail fluently too. Di third na responsibility. Wen system crash for 2am, somebody must carry di legal, financial and moral weight of di failure. No licence, contract or court dey assign dat weight to agent.
Labour market data dey tell di same story from di other side. Demand dey shift, e no dey collapse. Job postings dey increasingly ask for engineers wey fit direct AI tools, review generated code and design systems wey include non-human contributors. Di World Economic Forum future-of-jobs analysis dey always place AI and machine learning specialists among di fastest growing roles, even as e dey expect routine programming to automate. Di pyramid dey narrow for bottom and dey wide for top, na wetin dey happen to any profession wen im entry rung mechanise.
Real risk dey inside dis shape, and e deserve honesty instead of comfort. Senior engineers bin dey form by doing junior work slowly and badly. If machines swallow junior work, di apprenticeship pipeline go break. Companies wey dey celebrate cheap code today fit discover five years from now say nobody remain wey fit review di code wey dia agents dey write. Di rational answer no be to ban di tools but to redesign di ladder: pair juniors with agent output deliberately, make review a taught skill, and treat supervising machines as di new craft instead of confession of decline.
So, di title of software engineer go exist for 2028? Almost certainly yes, because titles dey die slowly and institutions dey conservative. Hospitals still dey employ doctors centuries after di role stop to mean wetin e bin mean. But di title go describe different daily reality: less typing, more deciding; less writing, more reading; less producing, more answering for wetin dem produce. Di lesson of di floppy disk no be say things dey disappear. E be say di object dey fade while di need survive, and na people wey follow di need dey prosper, no be people wey dey defend di object.
For leaders, di action concrete. Map which layers of your engineering work be transcription and which be judgment, den hire and train explicitly for di judgment layers. Build review capacity before you need am, because agent-written code still dey demand human wey understand am. And watch di apprenticeship question inside your own organisation before market ask you. Di title go take care of itself. Di pipeline no go take care of itself.
Think am mathematically. Engineer am well-well. Lead with strategy. Build am responsibly. Scale am globally. Use artificial intelligence create real business value.
