Observe reality
Research starts with what is breaking in the real world: participants, companies, agents, hiring, deployment, cost, latency, and user behavior.
How We Research
Teaching without research becomes repetition.
AI changes too quickly for static slides. The curriculum is built from continuous research, implementation, testing, and revision.
Antern research combines first principles, historical context, academic papers, industry validation, open-source implementations, production constraints, experiments, and feedback from real users and businesses.
Research Stack
The curriculum is updated from multiple kinds of evidence.
AI changes too quickly for a curriculum to depend on one source of truth. Antern combines academic literature, production systems, open-source code, failures, and business workflows.
Study history
New ideas are taught through what came before them, why older approaches failed, and what constraint forced the invention.
Read papers
Participants and instructors read papers for assumptions, methods, results, ablations, limits, and what the field learned.
Build systems
Research becomes useful only when it survives implementation: APIs, evals, traces, data quality, retrieval, agents, and production constraints.
Research Flywheel
Research becomes curriculum only after it survives the loop.
The loop turns raw information into teaching: observe reality, study history, read research, build systems, test assumptions, teach, and revise.
Observe
Find the real bottleneck in a workflow, model behavior, product system, or participant misunderstanding.
Study History
Ask what came before, why it failed, and why the current method became necessary.
Read Research
Read papers, surveys, engineering blogs, benchmarks, source code, and production postmortems.
Build
Implement the core idea or system pattern so the abstraction becomes concrete.
Test
Run experiments, evals, slices, failure probes, and user-facing checks.
Teach
Convert the research into a lesson, build, explanation, or curriculum update.
Revise
Use participant feedback, failures, and market evidence to update the curriculum again.
Paper Club
Participants are trained to read papers critically, not reverently.
Paper reading is not a prestige ritual. It is a way to learn research taste: what mattered, what failed, what changed, and what can be rebuilt.
Read
Abstract, introduction, figures, method, results, limitations, and related work.
Critique
What assumption does the paper make? What changed since publication? What did it miss?
Reproduce
Implement the core insight in code or recreate the essential experiment at a smaller scale.
Present
Explain the idea, why it mattered, what failed, and whether it still matters today.
Evidence Sources
Research includes papers, code, failures, and real workflows.
The most useful curriculum updates often come from the intersection: a paper reveals the idea, open source reveals the implementation, production failures reveal the limits, and teaching reveals what participants misunderstand.
Foundational research papersSurvey papersOpen-source implementationsEngineering blogsBenchmarks and eval reportsProduction incidentsParticipant build failuresEnterprise workflow constraints
What We Avoid
Research-driven teaching is the opposite of recycling.
The research process is designed to prevent stale teaching, tool worship, and hype-driven curriculum decisions.
- Teaching old slides because they are convenient
- Treating papers as scripture instead of arguments
- Teaching tools without explaining the constraint they solve
- Ignoring production failures because the demo worked
- Chasing hype without studying what actually changed
- Confusing confident AI output with research judgment