The Daily Work
Strip away the hype and a prompt engineer's week looks like disciplined applied engineering. They are not typing clever incantations into ChatGPT. They are running experiments against datasets, shipping versioned prompts through CI, and arguing with product managers about edge cases.
A typical day includes:
- Prompt design: writing and revising system prompts, few shot examples, and structured output schemas. This is maybe 20 percent of the job.
- Eval suite construction: curating test cases from production logs, labeling outputs, building golden datasets, and writing grader functions. This is the actual work.
- A/B testing: running prompt variants in staging or shadow traffic, measuring win rates on custom metrics, and shipping the winner behind a flag.
- Cost and latency optimization: moving work to smaller models where possible, adding prompt caching, trimming context, and negotiating with product to reduce scope.
- Incident response: when a prompt regresses in production, the prompt engineer is the one digging through traces to figure out why GPT started refusing benign requests.
The unsexy truth is that great prompt engineers spend more time on eval infrastructure than on prompts themselves. The prompt is easy. Knowing whether it is better than the previous version is hard.
The Tooling Landscape in 2026
The prompt ops stack has matured considerably. In 2024 everyone was rolling their own harnesses. By 2026 there are clear category leaders:
- LangSmith: the default tracing and eval platform for teams already using LangChain. Deep integration with agent workflows.
- Promptfoo: open source, config driven eval. Strong for CLI workflows and CI integration. Popular with smaller teams and indie developers.
- Humanloop: enterprise focused prompt management with collaboration features for product and engineering. Strong at prompt versioning and human review workflows.
- Braintrust: eval focused platform with good support for custom graders, dataset management, and experimentation across models.
- OpenAI Evals: open source framework from OpenAI. Useful for standardized benchmarks but less polished for custom production needs.
Most serious teams use two of these in combination, or pair one with internal tooling. The choice matters less than the discipline of actually running evals on every prompt change.
2026 LatAm Salary Ranges
Fully loaded annual compensation for remote LatAm prompt engineers in early 2026:
- Mid level (2 to 4 years, production ownership of prompt systems): $70,000 to $110,000 USD
- Senior (4+ years, eval infrastructure and cross functional leadership): $110,000 to $150,000 USD
You will notice there is no junior band. That is intentional. In 2026, entry level prompt work is folded into generalist AI engineer roles. Hiring someone specifically as a junior prompt engineer is a category error.
Salaries for pure prompt engineers trend lower than equivalent AI agent or LLM engineer roles, reflecting both the commoditization of the core skill and the narrower scope. If you are paying $150k for a senior prompt engineer in LatAm, you should expect them to own the entire eval and prompt ops stack, not just write prompts.
The Debate: Is Prompt Engineering a Real Role?
This is the argument that has not stopped since 2023. There are two camps.
The skills are becoming table stakes for AI engineers. The title is becoming a dated artifact of the early LLM era.
Camp one: the title is dying. Anthropic, OpenAI, and most frontier labs stopped hiring dedicated prompt engineers by late 2024. Their argument is that as models get better at following instructions, the marginal value of prompt expertise shrinks, and the work collapses into general AI engineering. Sam Altman said roughly this publicly in 2024. Karpathy has made similar noises. The job posting data backs it up: dedicated prompt engineer roles dropped sharply in 2025 while AI engineer roles grew.
Camp two: it is specializing into prompt ops. This camp points to large enterprises with thousands of prompts in production, where managing, versioning, and evaluating prompts is a full time specialty. In this view, prompt engineering is becoming what SRE is to software engineering: a discipline with its own tooling, practices, and career path.
Our take: both are right, for different segments of the market. At startups and product teams under 50 engineers, the dedicated prompt engineer role is fading. The skills are becoming expected of every AI engineer, the same way basic SQL is expected of every backend developer. At large enterprises with regulatory requirements and massive prompt surfaces, prompt ops is a real specialization that will persist.
If you are hiring in 2026, we would push you toward hiring an AI engineer with strong prompt and eval skills rather than searching for someone with "prompt engineer" on their resume. The talent pool is larger, the skill ceiling is higher, and the career path is clearer for the candidate.
What to Look For When Hiring
Whether the title is prompt engineer or AI engineer, screen for these concrete signals:
- Shipped eval infrastructure. Ask them to walk through an eval suite they built. If they cannot describe graders, datasets, and metrics specifically, they have not done the work.
- Experience with multiple model providers. Prompts that work well on GPT often fail on Claude and vice versa. Candidates who have only ever worked with one provider have blind spots.
- Cost and latency instincts. Ask them how they would cut a prompt's cost by 50 percent. Strong candidates immediately reach for caching, smaller models, and context trimming.
- Debugging war stories. Every real practitioner has a story about a prompt that regressed silently in production. Lack of such stories suggests lack of production experience.
- Opinions on structured output. JSON mode, function calling, and constrained decoding are table stakes. Candidates should have opinions about when to use each.
Key Takeaways
- Prompt engineers design prompts, build eval suites, run experiments, and optimize cost and latency in production LLM systems.
- The 2026 tooling stack is LangSmith, Promptfoo, Humanloop, Braintrust, and OpenAI Evals. Most teams use a combination.
- LatAm salary bands in 2026: $70 to 110k mid, $110 to 150k senior. There is effectively no junior band.
- The dedicated prompt engineer title is fading at frontier labs and startups, while prompt ops is specializing at large enterprises.
- When hiring, optimize for AI engineers with strong prompt and eval skills rather than searching by the prompt engineer title specifically.
Frequently Asked Questions
Is prompt engineering still a real job in 2026?
It is a real skill set, but increasingly not a standalone job title at most companies. Frontier labs like Anthropic and OpenAI no longer hire dedicated prompt engineers. At large enterprises with massive prompt surfaces, the specialization persists under titles like prompt ops engineer or LLM reliability engineer.
What is the difference between a prompt engineer and an AI engineer?
AI engineer is the broader role: they build production systems with LLMs, including prompts, agents, RAG, evals, and infrastructure. Prompt engineer is a narrower specialization focused specifically on prompt design and evaluation. In 2026, most prompt engineer responsibilities are absorbed into the AI engineer title.
How much does a prompt engineer make in LatAm in 2026?
Mid level roles range from $70,000 to $110,000 USD annually, and senior roles from $110,000 to $150,000 USD. These are typically 40 to 55 percent below US market rates for equivalent skill, and trend lower than comparable AI engineer or agent developer roles.
Should I hire a prompt engineer or an AI engineer?
For most teams, hire an AI engineer with strong prompt and eval experience. You get a broader skill set, a larger talent pool, and a more sustainable hire as the prompt specific title continues to fade. Only hire explicitly for prompt engineering if you have a dedicated prompt ops team and a clear career ladder for it.
Hire Prompt Engineering Talent with South
South sources LatAm AI engineers with deep prompt design, evaluation, and production experience, not just ChatGPT users with a fancy title. Tell us what you are shipping and we will have candidates in your time zone on a call within a week. Start hiring with South.
