When the AI costs more than the coder

The price of letting an AI write your code is rising fast, and by 2028 it could rival what you pay the developer sitting next to it – sooner in lower-cost markets than in Silicon Valley.

That is the warning from research firm Gartner, which predicts that AI coding costs will approach or surpass the average developer’s salary within two years, driven by surging large language model (LLM) token consumption and a shift to consumption-based pricing.

“Organisations are rapidly moving from experimentation to scaled deployment of AI coding agents, but many are underestimating the financial impact of rising token consumption,” said Nitish Tyagi, senior principal analyst at Gartner.

“Token discipline will not emerge through developer choice alone, as developers tend to optimise for speed and convenience over cost efficiency. Without a governed engineering operating model, costs can escalate faster than the productivity gains these tools are designed to deliver.”

Tokens are the units of data that generative AI models process, and they are the meter that AI coding tools such as Claude Code, Cursor and OpenAI Codex now run on. As vendors move from flat per-seat licences to billing by consumption, the cost of software engineering has become far harder to predict.

Gartner Peer Insights data, cited in coverage by Computer Weekly, shows 23% of tech leaders are already spending between US$200 and $500 (about R3 300 to R8 300) per developer per month on tokens, while 6% of organisations are paying more than $2 000 (about R33 100) per developer per month.

Expensive tokens

Tyagi told The Register that AI coding bills were leaping from $20 or $100 to $2 000 or $5 000 per developer per month (roughly R33 100 to R82 750), with extreme cases hitting $20 000 (about R331 000) in token charges. One such case, Tyagi told Computer Weekly, was an Indian IT organisation where a single developer’s AI usage was costing $20 000 a month, although the spend was ultimately justified because the work involved a complex legacy modernisation project.

Crucially, Tyagi said the prediction differs across markets. “We’re not saying AI token costs will be higher than every developer’s salary on the planet, because US salaries tend to be higher than in India, for example,” he said, noting that in India token costs may already be equivalent to the salary of an engineer with four to six years’ experience.

Read: South Africa risks losing a generation of developers to AI

South African developer salaries are closer to India’s than to Silicon Valley’s.

According to OfferZen’s 2026 Developer Salary and Benefits Report, based on more than 2 270 valid responses, a backend developer with under two years’ experience earns an average of R20 416 a month, and the average across junior developers with two to four years’ experience is R33 414. Intermediate full-stack developers with at least four years average R46 256, senior backend developers with six to 10 years earn R70 543, and the most experienced developers reach roughly R84 000 to R105 000 a month, with Cape Town paying the most at the senior end.

At the current R16.55/US$, the $2 000 a month that 6% of organisations are already paying (R33 100) is effectively a junior South African developer’s entire monthly salary. The $5 000 upper case (R82 750) rivals what a senior full-stack developer takes home, and the $20 000 outlier (R331 000) would cover three senior developers at once.

Gartner argues the core issue is not AI itself but a loss of cost visibility. The move to consumption-based pricing has introduced “highly variable cost structures”, and many vendors offer little transparency into how token consumption is calculated and billed, leaving enterprises unable to forecast or control spend.

“Most organisations still lack the maturity and frameworks to effectively measure cost versus business impact,” Tyagi said. “Software engineering leaders are increasingly concerned as token-driven AI spend becomes harder to justify, with budgets often being depleted earlier than expected.”

The firm points to common failure modes that quietly inflate bills: ungoverned autonomy in agent-driven workflows, bloated context windows and the absence of structured feedback to optimise usage. Vendors, it adds, have yet to ship mature, built-in cost controls.

To keep spend in check, Gartner recommends a disciplined operating model for software engineering teams: defining when agents should be used and at what level of autonomy; routing simpler tasks to smaller and cheaper models while reserving frontier models for complex work; training developers in context engineering to trim wasted tokens; setting token thresholds with automated monitoring; and folding token-usage reviews into sprint retrospectives.

Watching the meter

For all the cost anxiety, Tyagi is not telling teams to walk away. “I’m a big believer that AI is bringing gains. You should not move away from AI because the total costs are increasing,” he said. “But I believe that token costs will certainly increase.”