70% of Candidates Never Know They Faced an AI Interview. Dili's 9 Roles Are the Test.

The Nine Roles Dili Is Hiring For

Dili's LinkedIn post reported that Dili closed a $15 million Series A led by Khosla Ventures in July 2026, with Y Combinator, Brick & Mortar Ventures, and Allianz participating. In the six months since, LinkedIn's data shows the company has grown fivefold and now supports 700-plus of the largest energy and infrastructure projects in the U.S., tracking over $1.3 billion in infrastructure spend. The Y Combinator jobs board and BuiltIn list eight open positions as of September 2026. Five have detailed public postings; the remaining three appear in aggregate listings without full descriptions.

The Sale Operations Manager role is the most extensively documented. It sits between sales and compliance, owning vendor questionnaires, RFP responses, order forms, contract lifecycles, and the HubSpot pipeline. The posting asks for three years in sales or revenue operations at a B2B software company, hands-on HubSpot experience, and direct contract management, including redline work alongside legal. Preferred qualifications include exposure to renewable energy, construction, or regulatory compliance, plus familiarity with prevailing wage or Davis-Bacon concepts. First-year milestones are specific: own questionnaire and order-form workflows in 90 days, measurably shorten contract turnaround in six months, and build process that scales with headcount instead of breaking under it by year one.

The Product Manager role pays $160K–$190K with 0.1–0.2% equity, hybrid in Union Square four days a week. The first 90 days are mapped: learn the product and compliance domain by joining customer calls and shadowing internal teams; take ownership of a product area and lead discovery; ship a meaningful improvement and establish the roadmap. Qualifications emphasize shipping software used by real customers, strong product judgment, comfort with engineers on technical tradeoffs, and the ability to turn ambiguity into structure. Preferred qualifications include B2B SaaS experience, SQL or API familiarity, and early-stage startup background.

Senior QA Engineer and Demand Gen Marketer appear on the YC jobs board with location and, for QA, the same salary band and equity range. A LinkedIn post from June 2026 mentions a Customer Success Manager search tied to rapid growth and Fortune 500 customers. BuiltIn lists three additional roles: Compliance Customer Success Manager, Senior Full-Stack Engineer, Founding Business Development Representative, and DevOps Engineer. The pattern is clear: Dili is staffing the revenue engine (sales operations, demand gen, business development), the product core (PM, QA, full-stack, DevOps), and the customer interface (success) simultaneously. The compliance domain knowledge required for the sales operations role, including security questionnaires, vendor onboarding, and prevailing wage concepts, signals that the company's sales approach runs through procurement gauntlets as tough as the regulations its product automates. Candidates for these roles will face an algorithmic screen before a human ever sees their file.

Inside the AI Screen

The proxy problem is old. University name, past employer, and whiteboard performance correlate weakly with job success. Codility reports that 81% of companies use skills-based hiring, 94% say skills predict job success better than resumes, 70% of employers are shifting toward skills-first hiring, and 90% report fewer hiring mistakes with skills-based hiring. Fast-growing tech companies hiring at volume run candidates through platforms that replace proxies with observed behavior.

Codility has built a reference architecture. Its platform runs on nearly two decades of assessment science designed by occupational psychologists and recognized by the International Personnel Assessment Council. The same infrastructure serves GitHub, SpaceX, Tesla, EY, LSEG, SAP, Barclays, and Citi, representing 20,000-plus engineering teams in total. That infrastructure delivers two products: Codility Screen for hiring assessments and Skills Intelligence for workforce programs, both driven by a single assessment science model.

The candidate interface is deliberate. A person works in one screen: the task brief and source data on the left, an AI Assistant in the middle, the deliverable on the right. The task is a structured, real-world scenario with context and a decision to make against the clock. The candidate prompts, iterates, and builds a real deliverable — a verified report, a prioritization call, an account brief — alongside the AI Assistant. The output is a deterministic score, the deliverable itself, and the full trail of how AI was used. No generative AI scores the work. Experts define the answer space. The tasks cannot be completed without AI; that is the point. How a person works with AI is what gets measured.

The platform reads observable behaviors: prompting that produces reliable output, critical evaluation of AI responses, catching hallucinations and inconsistencies, iterating when the first result is poor, breaking down ambiguous problems, judgment on when to rely on AI and when to override it. These are the skills behind working with AI. The library covers customer support, success, sales, product, marketing, data analysis, finance, and general workforce tasks, growing monthly. Custom simulations can be built and scored against an answer space a company's subject-matter experts define.

For technical roles, the pattern extends into the interview itself. Codility's AI Copilot brings real-time AI chat and agent-based assistance to a VSCode-based environment. Interviewers see the candidate's AI interactions: AI Chat Mode for generic questions and code generation, Agent Mode for advanced workflows where the AI directly edits code and creates files, and auto-complete suggestions in the candidate's working space. Coming soon: interviewers will choose which models and modes the candidate can use. The platform is SOC 2 audited, ISO 27001 certified, CCPA and GDPR compliant, WCAG 2.2 AA accessible, and plugs into existing ATS and HRIS stacks with API access, and it is operational within four weeks.

Self-report rarely matches real skill. Completion rate shows who finished, not who can apply it. Manager opinion varies by manager with no repeatable yardstick. Multiple-choice tests read knowledge about AI; work simulations read whether someone can use AI to get work done. The platform's own figures show over 1.25 million candidate responses, with 87 percent rating the experience highly and fairness scores above 84 percent across every demographic group measured.

Dili's roles — spanning engineering, compliance, operations, and sales — would face different simulations from a library like Codility's. An engineering candidate might hit an AI-specific task built for real engineering roles, not generic coding puzzles with an AI wrapper. A compliance or operations candidate would face a business-task simulation where the deliverable and the AI trail are both scored. The screen is not a filter for who used AI; it is a filter for who used it well.

Getting Past the Screen

Workday's Paradox layer schedules tens of millions of interviews a year. HireVue has hosted over 70 million video interviews and 200 million chat engagements. Sapia.ai has processed 8 million interviews across 77 countries. The screen is not a gatekeeper you sneak past; it is a high-throughput parser you feed correctly.

The resume that passes is boring in the best way. Standard headings (Experience, Skills, Education, Certifications) in a single column. Dates, employer names, job titles, and locations exactly where parsers expect them. No text boxes, no multi-column layouts, no icons, no progress bars. If the portal preview scrambles fields, switch to DOCX immediately. A hiring manager at a midsize SaaS company that replaced HR specialists with an AI scanner put it plainly: "Your resume has to be simple, very basic, very boring. Words and phrases that match the job description. Metrics or results for most bullet points in experience."

HRLens analyzed the same pattern across Workday, Greenhouse, and similar systems: "If your CV clearly shows that a senior backend engineer used Python, AWS, Kafka, and led production systems, Workday, Greenhouse, and similar systems can match you to the role faster. If the same experience is buried under fluff like results-driven leader with a passion for innovation, recruiter bots and humans both miss the point." Your design does not get you interviews. Your proof gets you interviews.

The counterintuitive move is using AI to beat AI, but not the way TikTok suggests. Stuffing hidden keywords, copying the job description line for line, or dumping dozens of tools into a skills block fails. The workflow that works: extract the real requirements from the job description, map them to proof from your career (metrics, tools, team size, industry context, scope), rewrite only the weak or vague lines, then run the draft through a parser check for missing keywords, thin evidence, and formatting issues. Draft with one model, pressure-test with a second. The second model catches the first model's habits.

Consistency across surfaces matters more than optimization on any single one. Recruiters cross-check your LinkedIn profile, your application answers, and sometimes the first outreach message you send. If your CV says senior product marketing manager and your LinkedIn says growth storyteller, you have created noise for both AI resume screening and human reviewers. Use the same evidence base across all three. The goal is consistency, not cloning.

AI interviews and recruiter bots reward structure even more than resumes do. In a screening chat, a one-way video, or an automated phone step, give short answers with concrete evidence: situation, action, result. HireVue now sells AI hiring agents, text automation, video interviews, and assessments in one process. Sapia.ai uses structured chat interviews and explainable scoring; rambling hurts you more than short, relevant examples. If you are asked why you fit a customer success manager role, do not give a personality speech. Say you managed a $2.4 million book of business, lifted gross retention five points, and built playbooks that cut onboarding time by two weeks. Specifics travel well through transcripts, scoring rubrics, and human review.

OpenTrain's assessment data reinforces the same discipline: read the guidelines twice before the first task; most failures are guideline misses, not judgment errors. Rate substance, not style. Write specific rationales: "States the boiling point of water at sea level as 90°C; correct value is 100°C" beats "Inaccurate." Calibrate on the scored examples. Manage time: divide the limit by the task count. Do not guess at facts; graders reward honesty about uncertainty. Treat the first ten tasks as part of the interview.

The bigger play is building skills that stay valuable even when more screening becomes automated. Skills AI can't replicate are the messy human ones: prioritizing tradeoffs, running stakeholder alignment, handling conflict, managing change, explaining technical work to non-technical people, and spotting the real problem behind a vague request. AI can help you describe those skills better. It cannot hand you the lived examples recruiters trust. Your CV should show where you made decisions, aligned people, reduced risk, or changed an outcome that mattered.

The Junior-Rung Collapse

The data shows a fracture that began the month ChatGPT went public. A Harvard Business School working paper tracking 62 million workers across 285,000 U.S. firms from 2015 to 2025 found that junior employment grew steadily alongside senior roles until November 2022. Then the lines diverged. Junior hiring flattened, then fell. Senior employment kept climbing.

By the first quarter of 2023, headcount for early-career roles at firms adopting AI had dropped 7.7 percent after six quarters. In wholesale and retail — sectors where communication, documentation, and customer service automate cleanly — the decline hit 40 percent per quarter compared with non-adopters. Finance and technology posted the largest absolute reductions. A parallel Stanford study confirmed the pattern: U.S. employment for workers aged 22 to 25 in AI-exposed fields such as software development and customer service fell 13 percent relative to older cohorts, while experienced workers in those same sectors saw opportunities rise. Erik Brynjolfsson, director of the Stanford Digital Economy Lab and a co-author, put it bluntly on X: both his research and the Harvard study show "falling levels of employment for young workers in the occupations most affected by A.I. since early 2023."

The default use of AI is substitution: Let the machine do the work and cut headcount.

The mechanism is visible in the job postings themselves. The Harvard researchers analyzed nearly all U.S. vacancies from 2019 through March 2025. Occupations heavy on structured, repetitive tasks — the bread-and-butter of entry-level work — saw postings decline 13 percent. Roles demanding analytical, technical, or creative work that AI could augment grew 20 percent. The skill lists shifted too: automation-prone postings listed 7 percent fewer distinct skills, while AI-related requirements such as prompt writing and tool fluency appeared more often in augmentation-prone roles.

World Economic Forum research estimates that roughly half of typical junior tasks — report drafting, research synthesis, coding fixes, scheduling, data cleaning — can already be executed by AI. McKinsey puts the broader aperture at three in five occupations seeing at least a third of their tasks automated, though very few jobs disappear entirely. The squeeze is not replacement; it is compression. The same HBS paper notes that college graduates' lifetime wage growth depends heavily on early-career advancement that starts in low-paying entry roles. "If A.I. disproportionately affects junior positions, it could have lasting consequences for the college wage premium, upward mobility and income disparities," the authors wrote.

Employers are not blind to the pipeline risk. About half of senior decision-makers surveyed by Personnel Today said AI helps junior staff work better or faster, and one in four said it is creating new roles. LinkedIn reported in June 2026 that companies adopting AI rapidly were hiring more entry-level workers. But the aggregate data points the other way. The Harvard team warns that how firms integrate generative AI decides whether jobs vanish or evolve. Suraj Srinivasan, a co-author, recommends reskilling programs for automation-prone roles to build judgment and interpersonal communication, plus continuous upskilling in AI literacy for augmentation-prone ones. HBR's framework goes further: redesign tasks so juniors learn the why behind the work, not just the how; use "red teaming" exercises where new hires test AI outputs for flaws; embed them in workshops and client interviews where algorithms cannot reach.

The crunch is real. The screen Dili and others run is only the front edge of a labor market that has already moved the bottom rung.

Volume or Fit?

The argument over AI hiring has already moved past whether the software should exist. "So this is not an argument about whether the software should exist. That argument is over," wrote one observer analyzing Greenhouse's 2026 survey of 2,950 job seekers across five countries. "It is about being told the machine is in the room before we sit down in front of it." The numbers bear that out: nearly two-thirds of candidates have now faced an AI interviewer, up sharply in six months. Seven in ten were never informed beforehand. One in five only realized an algorithm was evaluating them after the interview had started. Fewer than one in five job seekers say their prospective employers have clear AI policies at all. More than half believe disclosure should be a legal requirement.

On one side sits the volume-first machine. Candidates are sending 50 to 200 applications in a single year, deploying automated agents to scrape job boards, track opening dates, and match exact keywords from job descriptions so their CVs clear automated screening filters. Tools like Sonara, LoopCV, and Gainrep's AI Auto-Apply search and apply around the clock. Simplify automates form-filling. The logic is advertising logic: more eyes, more chances. Marc Fiammante, founder of Gainrep, calls it a category error. "When AI is used to just create volume it strains the whole job market. The idea that finding a job is like advertising — the more eyes improve the chances — is completely wrong. High quality connections with real people who see the value in the person and the job is how interviews will always be determined."

Employers face the mirror image of that flood. Inboxes swell with AI-tailored résumés that increasingly "sound the same," as hiring teams report. Screening vendors promise throughput. But the Stanford Institute for Human-Centered AI has documented how the hiring AI pipeline — applications sent to vendor, machine learning models scoring, rankings returned — can yield racial bias and systemic rejection. Amazon's own experimental tool, scrapped in 2018, downgraded graduates of women's colleges and penalized résumés with the word "women's." In January 2026, job applicants sued to force disclosure of the "black box" logic behind algorithmic gatekeepers that block candidates before a human ever sees them. Iman Abuzeid, CEO of Incredible Health, said: "While AI tools may save time or help test candidates' skills, hiring managers shouldn't allow AI to make decisions."

The fit-first camp argues that AI's real advantage is bandwidth, not filtration. Autumn Gorman, head of product at Hired, framed it as an access play: "While teams cannot humanly interview every applicant, AI agents have no bandwidth limits. This gives every candidate a chance." Platforms are testing that premise. Hired inverts the marketplace: companies apply to pre-vetted candidates. Gainrep builds reputation-first endorsements, layering peer verification on top of the résumé. Adzuna's ApplyIQ markets itself as "responsible auto-applying." Jobscan and Rezi optimize for ATS passage but also make the matching criteria visible to the candidate. Teal functions as a command center, organizing applications across boards rather than spraying them.

The tension is structural. Volume-first tools optimize for the top of the funnel; fit-first tools optimize for signal at the bottom. Nina Alcorn, who tracks the ATS landscape, warns that "no two ATSs are the same. The technology changes fast, so an AI tool inside an ATS might work one way today and a different way tomorrow. Whether an ATS uses AI at all depends on the exact product and which features a team pays for and uses." That opacity is what the majority want regulated. Candidates describe countdown timers, digital avatars, and feedback like "lacked empathy" from an algorithm, which one report calls "uncanny and inhumane." The bleakness comes not from the screen itself but from the absence of a human nod, a smile, any signal that someone saw the person behind the keywords.

The debate won't settle on a single architecture. It will settle on disclosure, auditability, and the right to a human review, because the volume machine is already running, and the fit-first alternative only works if the machine tells you it's there.

What This Story Leaves Out

This article examines Dili's open roles and the AI-driven screening candidates will face. It does not cover the company's product roadmap, its internal culture, or the mechanics of its compliance platform beyond what hiring context requires.

Dili builds an AI-powered operating system for compliance in the built world. The platform ingests certified payroll, validates wage rates against live government data, and flags issues before auditors do, cutting review from seven-plus hours to under five minutes. Contemporary AI models operate only in the data layer, translating unstructured documents into structured data. A deterministic system then sorts that data against complex-but-static compliance rules. Fortune 500 clients building some of America's largest energy and infrastructure projects use Dili to automate prevailing wage and apprenticeship compliance. Roughly half those projects run Dili as an in-house software tool; the other half outsource the entire compliance process to Dili on a contractor model. CEO Anand Chaturvedi anticipates the industry will shift toward the software model over time.

None of that product architecture, customer traction, or business-model evolution appears in the sections above. The 15-person team size, New York City location, and Y Combinator Summer 2023 batch membership anchor the facts, not cultural analysis.

The competitive landscape, including legacy firms still running payroll out of QuickBooks, the sampling-based compliance model Dili replaces, and the "AI for compliance" pitch common among startups, stays in the background. The strategic wedge of Davis-Bacon and PWA compliance, the tier-two and tier-three subcontractor risk where misclassification lives, and the weekly statement of compliance somebody still has to sign are the market forces driving the hiring need. They are not the story here.

What remains is the hiring funnel itself: multiple roles, an algorithmic screen, and the tactics that move a candidate from application to offer. The same screen that filters for AI fluency at Dili is filtering out the junior rung across the economy. The candidates who learn to feed the parser today are the ones who will decide whether the machine stays a gatekeeper or becomes a gateway.

Working in AI? Zero G Talent tracks the openings: see every open Databricks role, browse AI jobs, openings at Anthropic, and the people building the field.