The stack is only as good as the program discipline around it.
In an agentic world, velocity is cheap. Defensibility is not. Teams that attach a chat window to every workflow will move fast and review poorly. Teams that place bounded agents inside clear requirements, security gates, and human sign-off will compound.
This is not a shopping list. It is how Stratir organizes the tools that sit behind research programs: what each layer is for, when to reach for it, and what must never be delegated to automation alone.
Three layers
Think, ship, and present. Most programs touch all three.
01
Think
Models and agent surfaces
Where requirements become code, research, and scoped agent sessions. The layer for exploration, implementation, and tool-routed work under operator control.
02
Ship
Security and operating rhythm
Where agent output meets review, issue tracking, and release discipline. The layer that keeps velocity from outrunning accountability.
03
Present
Design and narrative
Where research becomes interfaces operators can use and stakeholders can read. The layer that makes field releases legible, not just functional.
Think
Models and agent surfaces
Use these when translating requirements into implementation, running scoped agent sessions, or evaluating model behavior under program constraints.
Cursor
Cursor.com
Primary engineering surface
Daily development with codebase context, multi-file edits, and IDE-native agents tied to the repository.
Codex
OpenAI
Agent-assisted implementation
Multi-step refactors, test generation, and implementation passes that need a strong general coding model.
Hermes
NousResearch
Session agents and tool use
Local or scoped agent loops, tool-routed workflows, and research builds that need inspectable agent sessions.
OpenAI
AI Research
Models and APIs
Evaluation baselines, API-backed features, and model selection for production agent paths.
Pi.Dev
Inflection AI
Conversational building
Rapid prototyping and exploratory builds where conversational iteration beats formal IDE flow.
Ship
Security and operating rhythm
Use these when agent-generated code must pass review, security checks, and program tracking before it reaches operators.
Semgrep
Code Security
Static analysis
Security gates on agent output, policy checks in CI, and catching risky patterns before merge.
Macroscope
Code Review
AI-assisted review
Diff understanding, review acceleration, and second-pass scrutiny on agent-written changes.
Linear
Issue Tracking
Program rhythm
Scoping research tasks, tracking evaluation milestones, and routing work between humans and agents.
Present
Design and narrative
Use these when a release needs interfaces, motion, and narrative clarity—not only working logic.
Framer
Web Design
Marketing and product UI
Landing surfaces, rapid UI iteration, and narrative pages that must read as institutional, not templated.
Paper
Agentic Design
AI-native design
Exploratory UI generation and design iteration where agents participate in the layout process.
Spline
3D Design
Spatial assets
Visual product moments, spatial UI elements, and marketing assets that need depth without heavy engineering.
Rive
Animation
Interactive motion
Micro-interactions, product polish, and motion that communicates state without adding logic complexity.
Operating principles
How Stratir uses this stack inside research programs.
Agents inside scope
Every agent session has a defined task, tool limits, and a human owner. Open-ended chat is not a program.
Review before release
Semgrep and Macroscope are not optional polish. They are how agent output earns merge authority.
Track the program
Linear holds the evaluation path visible: what was requested, what shipped, and what still needs field proof.
Design for operators
Framer, Paper, Spline, and Rive serve field releases. If operators cannot read the surface, the research did not transition.
Build with discipline, not just speed.
The agentic era rewards teams that treat models as execution layers inside programs—not replacements for requirements, review, or operator judgment.
Discuss a program