← Back to Tracker
Milestone: First Production Client Confirmed
Vivid Windows
First real client for AI-built SPIFF
2 Agents
Desktop tester + org cleaner built
80-90%
Collection stage complete
7 Stages
Delivery framework expanded
Delivery Framework (Expanded to 7 Stages)
Original 5 stages expanded with 2 refinement stages. Percentages per Max Shevchenko's estimates.
Speaking Time by Participant
Mohammad Alazzouni
~10 min
Conversation Timeline
0:005:0010:0015:0020:0025:0029:00
Participants
Mohammad Alazzouni (delivery lead)
Max Shevchenko (SPIFF config, agent builder)
Taras Khudoba (facilitator)
Max Votek (strategy)
Maxim Solovyov (ops)
Key Takeaways
- First real client confirmed: Vivid Windows. Through Salesforce paper. Legal review underway with Roman. Client call scheduled for Tuesday to discuss AI approach (data, security, build).
- Two sub-agents built by Max Shevchenko: (1) Desktop tester — opens SPIFF, checks screens, validates designer. (2) Org cleaner — resets SPIFF org to clean state. Both tested and working.
- Rebuilt with new guardrails — no broken files. Using Mohammad's guardrails MD file + Jaspal's order-of-operations doc. Clean build on Taras's org. The configuration sequence issue from Apr 30 is resolved.
- 3 core knowledge files established: (1) Intake guide, (2) Build guardrails MD, (3) Order of operations (by Max Klyuchnikov). These form the context library for all AI agents.
- 40-step granular plan with rollback at every step. Full documentation generated automatically during configuration. Sanity test after each step ensures safety.
- Stage progress quantified: Collection 80-90%, Validation 90%, Configuration 80% (prototype working), Documentation 15-20%. Key unknown: production may differ from sandbox.
- Jaspal training session happened. Key insight: formulas are the trickiest part — "hundred ways to do it." Some things require SPIFF support tickets (e.g., import formatting).
- Agent architecture emerging: Consultant agent (client-facing, collects requirements) → Development agent (produces implementation plan, configures SPIFF). Human expert validates between steps.
- Framework expanded to 7 stages: Added "Refine Collection" (between 1 and 2) and "Refine Configuration" (between 3 and 4) — based on lesson that AI iterations get messy, sometimes better to restart from scratch.
- Valeria already using Excel generation on live projects — still needs human cleanup but producing usable artifacts.
- Maxim Solovyov: collect all deliverables per stage into one repo. Taras committed to scheduling artifact collection sessions.
- Max Votek proposed visual configurator — Cloudflare-hosted tool to help customers visualize compensation schemas. Audience segmentation needed (business user vs admin).
Score Trend Across All Meetings
Adopt
KB
Wflow
UAT
Roles
Biz
PM
Agent
Accel
Avg
vs Apr 30
+1
+2
+3
+1
--
+1
+2
+3
+1
+1.6
Workstream Assessment
1. Team Adoption +1
People actively using AI tools
5/10
Not started
Early
In progress
Mature
Jaspal training session completed. Valeria using Excel generation on live projects. Max Shevchenko building agents autonomously. Mohammad coordinating with Jaspal for guardrails. But Kevin and broader team adoption not discussed.
2. Knowledge Base +2
MD files, GitHub repo, reuse
6/10
Not started
Early
In progress
Mature
3 core files now established. Context library concept articulated. Guardrails MD (Mohammad), order-of-operations (Max Klyuchnikov), intake guide. Max Shevchenko's agent context folder has 5+ lessons-learned files. Solovyov pushed to collect all deliverables per stage into one repo. Taras to schedule collection sessions.
"Context library is the core of your knowledge based on which you can build different things" — Max Shevchenko
3. AI Workflows +3
Doc, config, test, implementation
7/10
Not started
Early
In progress
Mature
Biggest mover. Config workflow rebuilt with guardrails — no broken files. Desktop tester and org cleaner agents working. The "black box" problem from earlier meetings is being addressed: 40-step plan with documentation at each step. Import formatting issue is the main remaining technical blocker. Ready for production validation next week.
"Today it works fine, no broken files. The agent is already 100% ready for clicking, importing, API managing." — Max Shevchenko
4. UAT & Refinement +1
Feedback loop compression
5/10
Not started
Early
In progress
Mature
Refinement stages formally added to framework. AI iterations get messy — restarting from scratch sometimes cleaner. Key learning captured. Client call on Tuesday will be the first real UAT-style interaction. Rollback capability at every step provides safety net.
"Between one and two, refining the document based off of feedback was hard with AI — it kept getting messier, easier to just do it from scratch" — Mohammad
5. Role Compression = 0
Fewer roles, broader coverage
5/10
Not started
Early
In progress
Mature
Jaspal leading build, Max building agents, Mohammad coordinating — multi-hat pattern continuing. Vision of consultant agent replacing some discovery work discussed but not yet tested.
6. Business Model +1
Fixed-price / outcome-based shift
5/10
Not started
Early
In progress
Mature
First real client (Vivid Windows) will test the model in practice. Legal review with Roman underway. Through Salesforce paper — validates the channel partnership model. AI approach discussion with client on Tuesday will reveal client appetite for AI-delivered project.
7. PM & Tracking +2
Profitability, plan vs actual, forecast
4/10
Not started
Early
In progress
Mature
Solovyov's deliverables-per-stage push is a PM improvement. Taras scheduling artifact collection. 7-stage framework with % completion per stage gives structure. Max Votek's "where are we?" question forced quantification. Still no financial tracking.
8. Agent Architecture +3
Specialized agent experiments
5/10
Not started
Early
In progress
Mature
Biggest mover (tied). Two working agents built. Multi-agent architecture articulated. Desktop tester agent (login, screen checks, designer validation). Org cleaner agent (reset SPIFF org). Consultant agent → Development agent pipeline described. Regression testing agent planned. This jumped from concept to working code.
"We're going to have front end to the client asking questions... files from consultant agent for development agent... which produces the plan and builds the SPIFF fork" — Max Shevchenko
9. Accelerators +1
Packaged products (e.g. territory mgmt)
8/10
Not started
Early
In progress
Mature
Moving from prototype to production. Vivid Windows is the first live test of the accelerator. Visual configurator proposed by Max Votek. Post-project AI assistant concept discussed (clients asking questions after close). Valeria already generating Excel artifacts for live projects. The whole delivery approach is becoming the accelerator.
Blockers & Risks
Import formatting issue — data imported as strings instead of numbers/dates. May need SPIFF support ticket.
AT RISK
Production vs sandbox unknown — configuration may behave differently on client's real org.
AT RISK
Legal compliance for AI access to client data — under review with Roman.
WATCH
Connector to external system is out of scope for AI — manual work still needed on Vivid Windows.
WATCH
Action Items
Client call with Vivid Windows — discuss AI approach (data, security, build)
Mohammad (Tuesday)
Complete legal review / statement of work with Roman for Salesforce paper
Mohammad
Begin production build on Vivid Windows org (target: next week)
Max Shevchenko / Jaspal
Resolve import formatting issue — contact SPIFF support if needed
Max Shevchenko
Schedule meetings to collect and document deliverables per stage
Taras Khudoba
Add refinement stages (1b and 3b) to official delivery framework slide
Mohammad
Explore visual configurator prototype for customer-facing schema visualization
Max Votek
Generated from meeting transcript — SPIFF/SPM AI Configuration Project