{width=700 height=381} In many applied-learning ecosystems, “AI literacy” is starting to look less like an elective and more like infrastructure: a shared layer of workflow knowledge (prompting, automation, evaluation, and integration) that makes other skills productive. That framing is familiar in university settings where collaborative platforms (e.g., institutional GitLab instances) shape not just what students build, but how they document, iterate, and operationalize work. A recent Borna News report on Hamyar Academy’s new training phase provides a useful case study of how a non-university provider is trying to formalize AI capability as a practical system for freelancers and small businesses in Iran—while also raising questions about evidence, governance, and instructional rigor. ## Key analytical takeaways * **From “tool tutorials” to workflow design (in principle).** The report emphasizes durable competencies—system thinking, multi-step prompting, and process automation—over specific tools. This is a sensible educational move, because tools change faster than mental models. The open question is whether the curriculum operationalizes these concepts through reproducible artifacts (templates, rubrics, versioned examples) or keeps them at the level of motivational framing. * **Learner segmentation is plausible, but outcomes must be differentiated.** Positioning business owners, teams, and freelancers as distinct audiences matches real differences in constraints and incentives. Still, these groups need different success metrics: a store owner may care about lead handling and support deflection, while a freelancer may care about delivery time, QA, and client communication. Treating “AI adoption” as one uniform outcome risks oversimplifying what effective transfer looks like. * **The critique of fragmented tutorial learning is directionally right, but incomplete.** The report argues that ad-driven, non-sequenced content (e.g., scattered video tutorials) undermines systematic skill-building because it lacks feedback loops and a coherent progression. That aligns with learning science: structure and assessment matter. However, the best open resources can still be valuable when paired with a syllabus, peer review, and a project pipeline—so the contrast should be read as a critique of _unscaffolded_ learning, not of open learning per se. * **“Support as infrastructure” is the most interesting claim—and the most testable.** The report highlights continuous updates, recurring live sessions, a 24/7 AI assistant trained on the instructor’s materials, and human escalation. Conceptually, this resembles software education environments where issue tracking, CI-like checks, and discussion channels wrap around learning content. The practical questions are: how is the AI support evaluated for accuracy and bias, how are edge cases escalated, and what privacy boundaries exist when learners share business data in support channels? * **Localization can be a strength if it remains transferable.** Tailoring examples to local business realities can reduce friction and improve relevance. At the same time, localization should not become “prompt recipes for one market.” The strongest programs teach portable abstractions (problem framing, data hygiene, evaluation, and iteration) while using local scenarios as concrete testbeds. {width=559 height=372} ## Diagnostic checklist * **Define target workflows first:** list 3–5 real tasks (e.g., support triage, campaign analysis, proposal drafting) and specify measurable outputs (time saved, error rates, conversion lift). * **Look for an evaluation loop:** does the program teach how to verify outputs, measure quality, and iterate—rather than trusting “good-looking” generations? * **Check for project-based transfer:** are learners producing artifacts that can be reviewed (prompts, SOPs, automations, datasets, decision logs) with clear rubrics? * **Audit the support model:** what problems are handled by AI support vs. humans, what is the escalation path, and how is incorrect guidance corrected and documented? * **Inspect data and privacy practices:** are learners warned against sharing sensitive customer data, and are safe alternatives (mock data, redaction workflows) taught? * **Assess update sustainability:** do “free updates” map to a transparent maintenance plan (versioning, changelogs, deprecated modules), or are they primarily a marketing promise? > Ethically, treating AI as “infrastructure” should not mean normalizing manipulative practices (deceptive personalization, automated spam, or opaque persuasion). The most defensible educational stance is to prioritize user value, transparency, and verifiable performance—so that automation strengthens trust rather than exploiting it. ference: Borna News report (Farsi): https://borna.news/fa/news/2281426/