OrgX research & field notes
Experiments, benchmarks, and operating lessons from OrgX. We publish the evidence, the failure modes, and what changed—not just the launch.
Latest report
Jul 22, 2026 · Open standard
Agent Work Receipt v0.1 is an open, account-free contract for portable evidence, deterministic integrity, and independent verification across agent runtimes.
Archive
Jul 29, 2026Field note
Braintrust and Basis made agent behavior testable. The next primitive makes the judgment cumulative: an open work receipt that binds a behavior verdict to authority, artifacts, cost, and human acceptance.
Hope Atina · 8 min read
Jul 16, 2026Economics
Recall infrastructure tops out at $249/month. Proof gets compared against the dispute, the audit, and the cancelled project. OrgX prices like the second list.
OrgX · 4 min read
Jul 10, 2026Research
OrgX-Bench V1.2 adds solution-equivalence audits, counterfactual twins, contamination burn rules, statistical precision, and public corrections before any frontier headline can ship.
OrgX · 3 min read
May 30, 2026Research
The OrgX autonomous initiative benchmark had a quiet rot problem: its judge panel pointed at models that no longer exist. We fixed it, re-ran the full catalog on current GA models, fixed a cost bug along the way, and tied the autonomy score to the gated-autonomy controls that now ship in OrgX.
OrgX · 6 min read
May 27, 2026Field note
Linear, Jira, and Notion all expose useful MCP surfaces. OrgX is taking a different bet: the fastest agent workflow is the one where one tool call becomes durable company memory, execution state, and proof.
OrgX · 6 min read
May 5, 2026Product
Continuity is the missing infrastructure for AI-native companies. OrgX is the operating system that holds the thread between people, agents, decisions, and time.
OrgX · 5 min read
Apr 17, 2026Research
We ran 136 tasks across 4 models × 4 orchestration cells × 3 dependent task sequences. Single-shot benchmarks structurally hide what agents cannot fake: cascading context. Here is the data, the surprises, and what we changed.
OrgX · 7 min read
Apr 13, 2026Field note
The first five minutes decide whether your AI tools share company memory or become six separate rooms with six separate amnesia. That is why we built OrgX Wizard.
OrgX · 5 min read
Apr 12, 2026Field note
The point was not volume. It was forcing the system to find the few ideas with enough pain, specificity, and visual tension to deserve production.
OrgX · 5 min read
Apr 11, 2026Research
The OrgX autonomous initiative benchmark now publishes generated artifacts, independent judgments, token-level costs, and the failures that still need human review.
OrgX · 7 min read
Apr 11, 2026Field note
AI content looks like slop when a model is asked to carry taste, memory, and QA by itself. The fix is not a better prompt. It is a content system.
OrgX · 5 min read
Apr 11, 2026Thesis
Wearing many hats is not a founder superpower. It is a coordination failure wearing a flattering name. The founder should not have to be the operating system.
OrgX · 6 min read
Mar 26, 2026Product
I kept opening new Claude sessions and typing 'let me catch you up.' OrgX MCP is the continuity layer that stops founders from manually carrying context between ChatGPT, Claude, Cursor, and the rest of the stack.
OrgX · 11 min read
Mar 12, 2026Research
How OrgX-Bench separates public mechanism runs from private headline evidence, measures benchmark error, and evaluates trustworthy initiative completion.
OrgX · 15 min read