Mission — the two numbers that matter this quarter
North star · AI Agents
≈135 / 180 h·mo verified agent hours
180 h/mo by end July → stretch ~250 h/mo by end August · hours measured, not estimated
✓ Code Review ~40 h/mo
Company Research ×2 → merging
Proposal Gen · Portcast Jr · AI Analyst β
◉ J5 gate: ≥3 teams build solo — all-hands ~w/o Jun 15
North star · ETA Models
+3.63 / +5.00 pp weighted-OTPA, T≤7d
Beat current prod by ≥5 pp on 0–7 days-ahead by end August, sustained 14 days
✓ ETA-to-port live +2–3%
✓ Rules deployed: pred_clip_to_carrier(0.75d)
◉ Transit Time A/B since Jun 9 · readout end-Jun
Port Stay next · ML Gateway v1 in build
Needs attention 0overdue + due within 7 days · sorted by urgency · click to open in ClickUp
June focus — roadmap items, live statushand-curated from both roadmaps · as of 12 Jun
🤖 AI Agents
harden what runs · unblock in-flight · ship self-serve
J1
Fix code-review agent noise
Gate: 2 quiet weeks on every MR — first in line, protects trust in all agents
in progress
Daniel
J2
Security Review Agent v1
Dependency-age check first; running on MRs + defense-in-depth note
in progress
Daniel
J3
Merge the two company-research bots
One middle-ground agent; retire slack-er + portcast-company-researcher
not started
Pitch · Andrew
J4
Enable freight-audit alert agent
Built — waiting on safety sign-off (alert-only vs auto-refresh)
blocked: sign-off
Daniel · Pitch/SJ
J5
Self-serve onboarding, per team
MCP v2 ✓ · agent builder ✓ · all-hands ~w/o Jun 15 · gate: ≥3 teams build solo
on track
Andrew
J6
Lock the agent KPI for Q2/Q3
180 → 250 h/mo + empowerment metric + measurement rules
open · Nidhi
Andrew
🚢 ETA Models
ship wave 1 · lock the iteration loop
J1
Transit Time model → production
A/B started Tue Jun 9 · 2–3 wks to readout · gate: MAE ≥5% better on held-out routes
A/B live
Andrew · Pitch · Stephen
J2
Deploy winning combination rules
pred_clip_to_carrier(0.75d) live — 74.85 wOTPA, +3.63 pp vs carrier
done · +3.63 pp
Pitch
J3
ML Gateway v1 — single config
9 surfaces / 3 repos → one JSON blob; design report done, build underway
in build
Pitch
J4
Port Stay investigation
Residual error is Transit Time or Port Stay — reproduce training, GCP endpoint
starting
Andrew
J5
Lock OTPA / MAE KPI targets
Q2/Q3 numbers into the KPI sheet — same blocker pattern as Agents J6
open · Urvesh
Andrew · Pitch
Task boardslive ClickUp state · refreshed daily · click any task to open it
Cleared — recent completionsclosed tasks from the rolling 90-day ClickUp log
Next up — July & August at a glance
JUL AI Agents — teams build their own
Y1Marketing wave — GSC agent → Clay tiering → Ads copy · gate: ≥2 agents run by Marketing
Y2CEO Co-pilot v1 — daily brief + account brief · gate: Nidhi uses it 2+ weeks
Y3RFI/RFP agent — extend Proposal Generator (Emma)
Y4Analytics-agent pattern — Claude Code × Snowflake (Adrish)
Y5Catalog discipline — row at build time, monthly hours readout
JUL ETA — container-level intelligence
Y1Container trust meta-model — P(carrier wrong >1d) · gate: +5 pp F1 vs hand rule
Y2Best-of-N combiner — P/S/C + clip + trust score · gate: +1.5 pp vs June stack
Y33+ parallel A/Bs — prove the Gateway promise
Y4Per-lane / alliance / corridor rules — YAML variants, +2.8 pp each
Y5Container diagnostics — recurring 1 d/wk (Adrish + Andrew)
AUG AI Agents — make the hours real
A1Self-serve KPI Analyst — @kpi over Snowflake+PostHog · gate: ≥10 users, cited SQL
A2Measured hours + $ per agent · gate: ≥70% of rows measured
A3Marketing analytics consolidation — GSC + Ads + GA4
A4ETA Monitor & Auto-Retrain — joint with ETA roadmap, build once
A5Quarter review — re-rank ~25 backlog ideas on evidence
AUG ETA — automate the loop
A1CI/CD auto-deploy — nightly train → MR with 5% candidate variant
A2Partial-info coverage — +10 pp container coverage, no OTPA regression
A3Quicktrain learned rules — per-alliance OLS to real traffic
A4Drift monitoring per variant — must land same month as A1
A5Terminal ETA review — keep / refresh / gate per port
Open decisions & top risks
🔓 Decisions waiting on someone
- Q2/Q3 KPI numbers — blocks BOTH roadmaps. Agents J6 → Nidhi · ETA J5 → Urvesh + Pitch. One conversation could close both.
- Freight-audit alerts: alert-only vs auto-refresh — Pitch/SJ sign-off unblocks Agents J4 (agent is already built).
- Agent-builder platform — Claude skills vs Prototypes MCP vs dedicated builder; decide from what teams actually use during J5.
- ETA Monitor split — Agents A4 × ETA A1/A4: agree the boundary before anyone builds monitoring twice.
- Port Stay: model vs rule — June J4 answers this; August coverage work depends on it.
- RFI agent: extend Proposal Generator vs new build — decide at Y3 kickoff.
⚠️ Risks being managed
- Andrew is the bottleneck (Nidhi's explicit concern) → co-build rule; J5/Y1 gates count team-built agents only.
- Code-review noise erodes trust in all agents → J1 first in line; 2 quiet weeks before new engineering agents ship.
- API access stalls the July wave (GSC, Ads, Clay, financials) → request access in June; file-drop pattern for sensitive data.
- Customer escalation on 0–7d ETA accuracy → J1+J2 chosen for fastest 0–7d lift; weekly OTPA-by-horizon readout.
- Auto-deploy could ship silent regressions → drift monitor lands the same month, 14 green days before rollout.
- Hours-saved numbers look inflated → measured-hours protocol (A2); measurement rules locked up-front in J6.