Battlecard — Aditi Kashikar · Accountability

Salesforce Lead Developer Documentation Writer · 45-min single-competency interview · Competency: Get it Done → Take Accountability. Aditi is a senior leader: she runs a global, multidisciplinary org of 35+ defining the content vision for AI app-dev platforms, scaled a Content Design function (design-led onboarding, faster time-to-value), is a key stakeholder to Product and Engineering executives, and spearheaded an AI transformation program operationalizing AI inside the content lifecycle. But she's scoring one thing: do you own outcomes end to end, take personal responsibility for results and mistakes, follow through, and refuse to deflect?

Why her background matters: your AI-tooling stories — the eval, the drift detector, the AI Council adoption push — are her own AI-transformation program writ small. Mirror her vocabulary (operationalizing AI, velocity, time-to-value, end-to-end experience) and she hears a peer in the work, not just a candidate.
The rule for this room: every STAR makes clear I owned it, I delivered it, and where it went wrong, that was on me — here's what I did. Say "I owned that," "that was on me," "I held myself to" out loud.

What the JD Says They're Buying (weave these signals in)

How This Interview Works

Spine — Aditi hears the Accountability half

"My innovation exists to serve the customer, and I scale it through other people. I build the tools and the measurement, then I get the team using them so the win compounds."
Aditi hears "I build the tools and the measurement" — I own the outcome and prove it.

Story Routing — vanity-metric leads; one theme per slot

The vanity-metric arc is THE PRIMARY — it covers both the failure question and the "you were wrong / tough feedback" question (one arc, two slots). Git-migration is the fresh second failure: the one you caught yourself, with the correction staying at your level. Sonal heard the eval + QuickRamp; Diana heard Data Management + SAGE.

QuestionStoryWhy it's the right slot
"A failure / didn't go to plan" · "you were wrong / tough feedback"Vanity-metric → eval pivot (Pinecone)The Primary owned adopting the industry's incomplete metric, rebuilt around real success; the failure and your best work are the same arc
"A mistake you caught yourself" / second failureGit-migration confidence gap (Splunk)Fresh self-caught, fixed at your level; mentoring + change-leadership evidence
"Accountable without authority"AI Council holdouts (Splunk)the holdout version, not just the easy win — 68→98% in nine days
"How do you define success"Eval harness, clean (Pinecone)the misstep does not re-surface here — just the harness and the data
"Disagreed with leadership"Declined-to-publish the agent score (Pinecone)owned the honest number over the flattering headline
"Real failure with consequences"Shipped-doc-error (Pinecone)error went live; owned what shipped, closed the gap (confirm details)
"Docs honest at AI speed"Docs-drift detector (Pinecone)"dev's 10x pace isn't my excuse"
"Owned a result start to finish"Data Management (Splunk)Stale ⚠️ Diana heard it — re-angle only
Backup: second self-caught missLinter-v1 too aggressive (Splunk)if Git-migration is already spent

Prep four: vanity-metric (primary), Git-migration (self-caught), AI Council holdouts (no-authority), eval harness clean (success). One theme per slot — don't let them all drift to the same moral.

Bio (Director, Accountability lens)

Streamlined — 30 seconds (default; lead with this)"I got my start as a technical writer at Workday over 11 years ago, then spent most of my career at Splunk, where I was a senior staff technical writer and also led teams and initiatives that operationalized AI, like our docs-as-code migration and an AI adoption push I ran from 68 to 98 percent. Then I pivoted to a high-velocity AI startup, Pinecone, to get into the weeds and really understand how AI works. There the question I own is simple: can an agent read our docs and finish the task? I built an eval harness that turns that into a number. My first attempt chased the industry's agent-friendliness score, which tracks discoverability, not success, so I rebuilt the measurement around task completion. I've run it against Salesforce's own docs. I own the outcome end to end, misses included, and that's what I'd want to own here."
Full — 60 seconds (if she wants more runway)"I got my start as a technical writer at Workday over 11 years ago, then spent most of my career at Splunk, through the Cisco acquisition, where I was a senior staff technical writer and also led the teams and initiatives that operationalized AI in how we worked: I took six writers from Google Docs to docs-as-code, drove a docs restructure from support ticket to shipped change in a week and a half, and ran our AI adoption push from 68 to 98 percent. Then I pivoted to a high-velocity AI startup, Pinecone, to get into the weeds and really understand how AI technology works. There I own whether docs actually make an agent able to use the product, and I've answered that with data: an eval harness that scores real agent task completion at four levels of documentation. I built it after my manager challenged my first attempt, which leaned on the industry's agent-friendliness score, and that only tracks discoverability, not success. I had an incomplete picture, so I rebuilt the measurement instead of defending it. I've run the real one against your Multi-Framework beta. I own outcomes, including where I was wrong, and that's the work I want to do here for the next five years."

⬛ When She Asks the Failure Question — own it in one sentence, then pivot to the fix

Open with the routing question: "Happy to. Would you rather hear about a mistake I caught myself, or one where someone else had to point out a blind spot?" It routes her to the right story (self-caught → Git-migration; blind spot → vanity-metric), buys a beat to set up, and signals you know mistakes come in both flavors. One sentence, then answer; if she says "either," default to vanity-metric.

The failure question is the room. The trap is burying the lede or over-explaining the setup so it sounds like you're dodging. The accountable move: name the miss cleanly and fast, then spend the answer on what you changed.

"Sure. Early in the docs-for-AI work, nobody really knew how agents used docs, so I chased what the industry was measuring, an agent-friendliness score that ranks companies' docs. I spent a week optimizing against it, built a dashboard of where we sat against other companies, and I was ready to market it. My manager asked one question: that covers discoverability, but what about success? He was right, I had an incomplete picture, and I rebuilt the measurement instead of defending it…"

Aditi — Get it Done: Take Accountability

Listening for: do you own outcomes end to end, take responsibility for results AND mistakes, follow through, and refuse to deflect? Every STAR: I owned it, I delivered it, and where it went wrong, that was on me.

Primary  The vanity-metric pivot (failure AND wrong/feedback — your centerpiece)

Covers both "tell me about a failure" and "a time you were wrong / took tough feedback." The failure and your best work are the same arc. Reference the eval in one line — Sonal heard the mechanics. Framing: the score is Fern's industry Agent Score (buildwithfern.com/agent-score), not yours — the miss you own is adopting the industry's metric as the definition of success without validating it. Incomplete, not wrong (discoverability ≠ success); own the incompleteness crisply, never defend the metric.

SWhen I started the docs-for-AI work at Pinecone, it was on me to define how we'd measure whether docs worked for agents, and honestly, nobody really knew how agents were using docs yet. I keep up with the industry, and what everyone was chasing was agent-friendliness, Fern publishes an Agent Score that grades companies' docs on retrieval signals, llms.txt, chunking, frontmatter. So I chased it too. I spent about a week optimizing our docs against it and shared a dashboard with my manager: our score, where we sat against other companies, and I was already thinking about marketing it, a blog post on how agent-friendly our docs were.
MomentMy manager asked one question: okay, that covers one thing, discoverability, but what about success? And he was right. The score tracks whether an agent can find our content, which is real, but an agent can find a doc and still fail the task. I'd basically adopted the industry's measurement as our definition of success, which was an incomplete picture, and I was about to make it our headline, internally and publicly.
AI didn't catch it myself, and I own that. I reprioritized on the spot: kept the agent-friendliness score as an input and rebuilt the measurement around the part that was missing, an eval harness scoring real task completion at four levels of documentation, no docs, a baseline file, live fetch, and a curated skill. Discoverability still feeds in, it just doesn't get to be the score.
RThe curated docs hit 100% task completion where live retrieval and no docs both hit zero, a finding the friendliness score alone would never have surfaced. The lesson: I'd spent a week chasing what everyone else measured instead of validating it against our outcome. Now I check that a metric predicts the outcome before I commit to it, so the catch happens at my level.
Line"The accountability was admitting my measurement was incomplete and rebuilding it around what success actually means for an agent: can it finish the task. And part of it is working in the open. I shared that dashboard after a week, not a quarter, so the pivot was fast and we didn't waste much time. I share work early and I check my metrics against real outcomes. Those are the systems that keep me accountable."

If asked "who's your manager / their title": answer "VP of Product" immediately and plainly — small org, docs lives under product. The only version that hurts is a pause or walk-back.

Strong · Fresh  Git-migration confidence gap (the mistake you caught yourself)

The second failure story, and the self-caught one — correction stayed at your level. Also your best mentoring + change-leadership evidence for the JD.

SAt Splunk I led my team of six writers from Google Docs to docs-as-code in GitHub.
TAdoption was on me, and I'd planned the rollout around mechanics: who submits PRs, who reviews, how handoffs work.
AIt stalled, and that was my planning, not the team. One writer wasn't resisting the idea, she just wasn't confident with Git itself, what merging does, the fear of breaking something. I'd built a process plan when the team needed a confidence plan. So I made short video walkthroughs of the exact clicks and commands for each path.
RAdoption recovered and we finished the migration. Now I plan for the skills gap up front, because adoption is the owner's job.
Line"The stall was my planning gap, not their reluctance. I caught it, closed it, and I've planned for the skills gap ever since."
Strong  AI Council holdouts (accountable without authority — 68→98% in nine days)

Use for "how do you hold people accountable without authority." The holdout version, not just the easy win. Dana's room gets this as Influence — here it scores Accountability.

SAs head of our AI council at Splunk, I owned a training completion target that had been stuck at 68% for three months. I had the responsibility but no stick, I couldn't mandate or incentivize anything.
THitting it meant moving people with no obligation to move.
AI went after the friction first: replaced a clunky screenshot-upload step with a demo video, got us a dedicated Friday block, kept our standing visible in Slack. The real holdouts needed a different conversation, so I made it concrete that engineers were already moving at this pace, and staying trained is how you stay relevant. And the genuinely stuck, I followed up with one on one.
R68% to 98% in nine days. The one person we missed was on PTO.
Line"No authority doesn't mean no ownership. I removed the friction, then worked the holdouts one by one. That number was mine to move either way."
Strong  Eval harness, clean version (how do you define success)

Pure win framing — the vanity-metric arc belongs to the failure/feedback slots; don't re-run the misstep here.

SAt Pinecone I owned whether our docs actually worked for AI agents.
TThe industry default was traffic, and traffic doesn't tell you whether an agent can do anything with the docs.
AI built an eval harness on Sonnet 4.6 that measures task completion at four levels of documentation: no docs, a baseline file, live fetch, and a curated skill.
RReal data on where docs moved the needle. I later ran the same approach against Salesforce's docs.
Line"Success is the agent finishing the task, and you can measure that. I'd rather own a pass-fail number than a traffic chart."
Strong  Declined to publish the agent score (owning the honest number)

Sequencing vs the vanity-metric arc: early on you were the one planning to blog about the score; post-pivot you're the one keeping it internal. If both come up, make the before/after order explicit.

SMarketing at Pinecone wanted to publish a rising "agent score" as a headline.
TI owned what we published, and I'd be the one accountable the day a customer tested that number.
AI kept it as an internal diagnostic and pointed the public story at real task completion. A number we couldn't defend would cost us trust.
RWe shipped a number we could stand behind.
Line"I'd rather own a small number that's true than a big one I can't back up."
Strong  Docs-drift detector (accountable to dev's speed, no excuses)
SAt Pinecone, engineering was shipping about ten times faster with AI. The easy out for docs is "I can't keep up," and then docs fall behind permanently.
FramingI don't get to use dev's speed as my excuse. If they're multiplying output with AI, I use the same leverage to keep docs correct. Docs staying true to the product is still on me.
AI built a drift detector: when code changes in one repo, it automatically opens a doc PR in the docs repo. But I'm still the reviewer in the loop.
RDocs keep pace, and a human still gates quality.
Line"'Dev moves too fast' is my problem to solve. The pipeline suggests the change, I own what ships."
Strong (real stakes)  Shipped a doc error my automation missed

The "real failure with consequences" story. ⚠️ Confirm the real details before you say it. Two traps: don't blame the tool, don't make the hero "before a customer saw it."

SI own the docs for a fast-moving SDK at Pinecone, and I'd built automation, the drift detector and the linter, as the safety net. I trusted it.
MistakeA doc went out with an error [name it: wrong parameter / stale sample / wrong limit]. My automation didn't flag it, my review didn't catch it, and a developer found it.
AFirst, I didn't blame the tool. I built it, so its blind spot is mine, and I own what ships, not the pipeline. I fixed the error right away. Second, a mistake caught downstream means my process had a hole, so I added a check for that whole class of error. And I took the catch from that developer straight, someone flagging my miss is doing me a favor.
RThe fix was fast, but what I'm accountable for is closing the gap. That category of error can't slip through silently anymore.
Line"I own what ships, including what my own tooling fails to catch. The real fix was hardening the system, not the quick patch."
Backup  Linter-v1 too aggressive (second self-caught miss, if Git-migration is spent)
SI built a CI style linter at Splunk and tuned the first version to catch everything.
ScarIt threw too many low-value flags and writers started tuning it out. A check nobody trusts is worse than none, and that was on me.
AI caught it myself, cut it to the high-confidence rules, and had the writers tell me which flags were noise.
RAdoption came back. It ended up running across a hundred-plus contributors and cut manual review from five hours to three.
Line"Good tooling earns trust by being quiet and right. I owned the miss and fixed the trust."
Re-angle only  Data Management owned end-to-end (follow-through and delivery)

⚠️ Diana heard this in full. Use only for "owned a result start to finish" if nothing else fits, and re-angle for Accountability; don't re-recite.

SA buried product area at Splunk was generating high-traffic support tickets.
AI took it from the ticket to a shipped fix. I pulled in the PM, another writer, and Support, but the chain holding together was on me: I set up the meetings, kept us on one goal, and drove the IA change that gave it its own tile. If it stalled, that was on me.
RShipped in about a week and a half, closed the ticket, roughly a 20% findability lift.
Line"I don't hand off a problem and hope. I own it to the shipped change and the closed ticket."

Likely Accountability Questions + Routing

First-90-Days POV (volunteer it — signals ownership)

🟢 The React-App Demo Offer — gauge her interest

QuickRamp is a real React app you built on Salesforce's Multi-Framework beta with a live /eval page. For Aditi it's pure Accountability: you didn't assert "the docs hold up for an agent," you built the product and proved it with data you own.

Say this: "And I don't just claim the docs work for an agent. I built a real React app on your Multi-Framework beta with a live eval page inside, so the claim is measured. Want me to put it on screen for two minutes?"

Questions to Ask Aditi — director altitude; signal you'd own a number, and that you've read her work

Watch-Outs

Cheat Sheet (if you freeze)