Anonymized portfolio copy — names and customers replaced; all data illustrative.
Adoption
Who is using AI. Measured natively across the org.
“Are people turning it on?”
→
Effective usage
Who is getting value from AI. Derived / estimated.
“Is it actually producing?”
Structural decision (01 obj. 5). The report keeps adoption and effective usage visibly separate everywhere, because the data already splits this way — adoption is measured, effectiveness is estimated. This green/amber split runs through every section below; it is not just a Section 2 device.
1
How is AI adoption impacting delivery?
One consistent story in three numbers: AI is already an Nx accelerator; it has saved real money; and there is more on the table. This is the line leadership should leave with — so it leads the report.
Data reality (02 §1). None of the three hero numbers exists as a served metric. The multiplier is a derived ratio whose denominator — a historic “traditional development” baseline — is not stored anywhere. Savings inherits that gap. “Potential savings” is a counterfactual the platform explicitly cannot capture. All three are red-bracketed here on purpose: the narrative is sound, the figures await decisions (Open Decisions 1–2). The trustworthy, measured foundation is Section 2 — built to earn the credibility this row borrows.
① Acceleration
[N]×
AI acceleration multiplier
▲ trend over last 6 months — [derived w/ formula]
= AI-era output rate ÷ historic “traditional development” baseline rate. Baseline not stored → Open Decision 1.
② Realized
$[X]
Savings realized from AI
▲ cumulative, last 6 months
= (baseline cost − actual cost), derived from the multiplier × engineering-cost components (estimated). Inherits the baseline gap → Open Decision 1.
③ On the table
$[Y]
Additional potential savings
[needs reframe — see note]
Counterfactual “spend avoided” is non-capturable. Honest alternative: the lagging-cohort efficiency gap (Section 3). → Open Decision 2.
✦ Takeaway (executive voice): “In one line: AI is making my engineering org [N]× faster, it’s already saved us $[X], and there’s $[Y] more if we close the gap.” — the single sentence this row must deliver once the figures resolve.
⚠ Risk on this row. If the three figures ship bracketed, the report opens on its weakest data. Mitigation built into the spine: Section 2 (measured) sits immediately below and visually outweighs this row until the multiplier/baseline decision lands.
2
What is the state of AI adoption across the org?
The measured foundation of the report. Read left-to-right as adoption (who’s using it) then effective usage (who’s getting value) — the report’s core structural line.
Adoption — measured
AI adoption rate Measured
Basis: AI-active engineers ÷ active engineers. Native rollup.
75%
of active engineers used AI this period
▲ rising 6-mo trend
Daily active AI usage Measured
Basis: daily AI-active ÷ active, effective data window.
68.8%
use AI on a typical day
▲ rising 6-mo trend
Power-user rate Measured
Basis: engineers above the heavy-use threshold (native).
62.5%
are heavy, sustained AI users
▲ rising 6-mo trend
✦ Takeaway: “Adoption isn’t my problem — three in four are on it, most of them daily.” — exec voice.
Effective usage — estimated
Share of code built with AI Estimated
Basis: retained AI-assisted share of retained lineage units. [demo 99.7% — implausibly high; real-tenant value differs]
[~%]
of retained output is AI-assisted
Value per dollar Estimated
Basis: value-weighted output ÷ total cost.
0.36vLOC / $
efficiency — the “effective usage” counterpart to adoption
Cost per outcome Estimated
Basis: total cost ÷ value-weighted output (vLOC).
$2.75/ vLOC
what one unit of value-weighted output costs
✦ Takeaway: “People are using it — but value-per-dollar is where I find out if it’s actually paying off.” — exec voice.
Budget lever (01 obj. 9)
AI-budget utilization Estimated
Basis: we can show AI spend and token count today; utilization against a budget has no backing fact (no budget/quota data) → Open Decision 4.
$175k
AI spend, this period est
avail.
token count — available
[ % of budget ]
utilization — needs budget facts
Token cost control is the company’s top priority — framed here as a lever to encourage, not waste to police (AI-waste detail is out of scope for this exec surface). Utilization-against-budget red-bracketed until budget facts land.
Why this section leads on trust. Adoption metrics are native rollups (measured); effectiveness metrics are derived (estimated) and badged as such. Keeping the two columns visually distinct is the on-page expression of obj. 5.
3
Where does the output actually come from?
Engineers ranked by efficiency (value-output per dollar), and the share of code the top cohort produces. Leads with efficiency over raw output (01 obj. 9).
Engineers ranked by efficiency Estimated
Basis: value-weighted output per engineer ÷ cost. Ranking sound; absolute values estimated.
| Cohort | vLOC / $ | Share of engineers |
|---|
| Top | 0.61 | ~25% |
| Middle | 0.34 | ~50% |
| Lagging | 0.18 | ~25% |
Demo values illustrative; Step 5 arithmetic-footing pass must reconcile cohort vLOC/$ against the org-wide 0.36 in Section 2.
Share of code the top cohort produces Measured-ish
Basis: retained output summed per engineer, ranked. Concentration is directly countable; AI/human split is estimated.
~60%Top 25%
~30%Middle 50%
~10%Lagging 25%
The top quarter of engineers produces the majority of retained output — the “what % of the team produces most of the code” answer leadership asked for.
Cohort bands: 3 vs 5 — Open Decision 5. Shown as 3 bands here; data supports either. Banding is a pure design choice over the ranking.
Honest “money on the table” (02 §3 reframe). The lagging-cohort efficiency gap — bottom-cohort spend × (top − bottom value-per-dollar) — is a knowable forward figure built from served components. This is the defensible replacement for the §1 counterfactual “potential savings.” Surfaced to the CEO/Trent as Open Decision 2.
✦ Takeaway: “A quarter of my engineers drive most of the output — and closing the gap on the laggards is the real money on the table.” — exec voice.
4
What sets the top cohort apart?
An aspirational signal: characteristics of the highest-efficiency engineers, so the rest of the org has something to move toward. Kept lightweight — partly data-dependent.
Adoption depth Measured
▲
Higher daily-active and power-user rates than the org average.
Usage intensity Estimated
▲
More prompts / sessions per retained outcome.
Tool & model mix Estimated
[?]
Leans on proxies / includes “unknown” — keep light or defer.
Feasibility (02 §3, 01 obj. flag). Adoption-style differentiators are defensible (native). Behavioral “how they prompt” differentiators lean on estimated proxies. This section stays aspirational and lightweight — or defers to a later cycle if the data can’t carry it honestly. No fabricated “top performers do X” claims.
✦ Takeaway: “Here’s what ‘good’ looks like — and it’s reachable for the rest of the team.” — exec voice, aspirational.
Not in this report (deferred or out of scope): Quality metrics — cycle time, completion-quality, review/defect rates — deferred under the AI-adoption frame (and the data backs deferral: those facts are unavailable). · AI-waste detail — out of scope for this executive surface; lives in the department/team view. · Supervised-vs-autonomous code split and true agentic-developer share — not captured today; roadmap, not v1.
Open decisions
Every red [bracket] above traces to one of these. Decisions, not data — carried forward to Step 4 / stakeholders. Resolve with name + date; never invent a number.
- 0 · Report rename — adopt “AI Engineering Adoption Report”? Session leans yes; not finalized. (masthead title)
- 1 · Multiplier formula + baseline source — no historic “traditional development” baseline is stored. Define the formula and the baseline, or the hero multiplier stays bracketed. Blocks the hero row and realized-savings. (§1 ① ②)
- 2 · “Additional potential savings” — counterfactual “spend avoided” is non-capturable. Reframe as the lagging-cohort efficiency gap (§3), or flag. Raise with the CEO/Trent. (§1 ③)
- 3 · “Agentic / GenAI developer” definition — usage-threshold (backed by power-user rate) vs. “% of code via agents” (unbacked — needs the autonomous split). Pick one explicit definition. (masthead unit def)
- 4 · AI-budget utilization — no budget/quota facts exist. Show spend and tokens only, or red-bracket utilization until budget facts land. (§2 budget lever)
- 5 · Cohort bands — 3 vs 5 bands. Pure design choice; data supports either. (§3)
- 6 · Quality metrics — confirm deferred under the adoption frame (data supports deferral). (deferred strip)