"SDLC Is Out, ADLC Is In": The Stockholm JD Breaking LinkedIn Search

A Series A in Stockholm posted an Engineering Director JD in the July 2026 HN "Who is hiring?" thread that opens with a line I've never seen in a real posting before: "SDLC is out, ADLC is in." The plan is 5 to 7 teams of 30 to 40 engineers where AI writes most of the code. The problem is that almost nobody who could actually do this job is findable by the keyword search a sourcer would run.

This post is the sourcing playbook for that JD, and for the wave of ADLC roles right behind it.

What ADLC means, and why it breaks SDLC sourcing

Agentic Development Lifecycle (ADLC) is a software lifecycle for systems where LLMs sit at the core of product behavior, not a version of SDLC augmented with coding assistants. IBM, Atlan, Arthur.ai, Cycode, Salesforce Agentforce, and Cognizant have each published formal ADLC frameworks, which means the vocabulary is about to spread from Series A JDs to Fortune 500 requisitions within 6 to 12 months.

Three things about ADLC break the way sourcers currently search:

• The unit of work changes. Per the agenticlifecycle.ai reference, "the unit of account is cost per merged, verified change. Not tokens per developer per month. This is the SDLC inverted, because misbuilding is expensive and building is cheap."
• The core artifact changes. Prompts, model choices, and eval thresholds get versioned as first-class engineering artifacts. Sumatosoft calls the resulting document an "AI Decision Record," a variant of the ADR tradition.
• The reliability work is the hire. Arthur.ai puts it plainly: "getting to a functionally complete state tends to be pretty quick. Going from functionally complete to reliable is where the work lies."

That last point is the whole game. The Stockholm JD is not asking for "Copilot power users." It is asking for people who have shipped an eval harness, owned a drift dashboard, and lived through a production agent outage. Those signals do not exist in LinkedIn's headline field.

The Stockholm JD, in numbers

The JD sits in HN item 48747976 alongside other Stockholm agentic-AI roles, including Pango (YC S26), a "Founding Full-stack Software Engineer" building an "Agentic Operating System for e-commerce logistics" with visa sponsorship on the table. The Engineering Director role sitting alongside it is what makes the thread notable: it treats ADLC as an org-design principle, not a buzzword.

Here is the math a sourcer sees when they take that JD at face value and search for its stated skill stack. Every count below comes from Refolk's index of professional profiles.

1

The Stockholm supply number is the punchline. Three people in Sweden with an AI or Agent Engineer title listing LangGraph. The JD implies staffing 150-plus engineers over time. The delta is not a recruiting funnel problem, it is a geographic reality that visa sponsorship is trying to paper over.

Why LinkedIn keyword search returns almost nobody

LinkedIn keyword search returns almost no qualified ADLC candidates because the qualifying work happens in repos, evals, and prompt versioning systems that never make it into a headline or a skills tag. The people doing this work are heads-down builders, not profile updaters.

Four specific reasons the recall is so bad:

1. The stack is new. ADLC as a term is under a year old in the JD wild. Claude Code and MCP are younger than most keyword filters' "years of experience" thresholds.
2. The vocabulary hasn't stabilized. Some teams say "agent lifecycle," some say "context engineering," some say "eval-driven development." No single phrase catches them all.
3. A large share of visible LangGraph talent are founders. In Refolk's index, 7 of the top 25 title matches for LangGraph across a 4-country search are CTOs or co-founder CTOs. Recruiters chasing LangGraph on LinkedIn are largely chasing people who won't take a job.
4. The real signal is negative. Per EPAM's ADLC framing, the lifecycle "is not SDLC augmented with AI coding assistants." Candidates whose headline brags about being a "Copilot power user" have failed the actual test.

The 15-person global pool for the exact skill stack is not because only 15 humans can do this work. It is because only 15 humans have chosen to advertise it that way. The rest are the exact gap Refolk closes: you describe the person in plain English ("engineers who've shipped a LangGraph agent to production with an eval harness") and get a ranked shortlist pulled from GitHub, LinkedIn, and the open web, not just LinkedIn headlines.

What to search on instead of headlines

Search for the artifacts, not the labels. The candidates you want have created specific files, repos, and commit histories that keyword-only tools cannot see.

Five signals that beat headline keywords for ADLC sourcing:

• evals/ directories in public repos. Arthur.ai calls a well-curated eval suite "the single biggest unlock to guarantee success in an Agentic System." Candidates who ship one leave a directory.
• Commit history in promptfoo, braintrust, or inspect-ai. These are the eval frameworks that separate hobbyists from production operators.
• MCP server repos. Anyone who has published a working MCP server has done more agent engineering than 100 people who list "MCP" as a skill.
• LangGraph subgraphs with checkpointing. Toy LangGraph demos are common. Persistent state, human-in-the-loop interrupts, and long-running graphs are not.
• Writing that names specific model versions and eval thresholds. "Tool-call accuracy went from 78% to 91% after we changed our eval gates" is a hiring signal. "AI is transforming software" is not.

Kadoa's July 2026 HN thread is a live example of the profile this surfaces. Applicant "princek" cited building agent systems (IntentFrame) and deep web-automation background - a self-taught agent builder whose LinkedIn headline almost certainly does not carry the JD's keyword stack, but whose repo history does.

The 30 to 40 engineer team size is an economic claim, not a headcount plan

The Stockholm JD's "5 to 7 teams of 30 to 40 engineers" is not a growth-at-all-costs headcount grab, it is a bet on ADLC economics where review is the bottleneck instead of authorship. That inverts a decade of startup wisdom about small elite teams.

The mechanism is simple. If cost per merged, verified change is the unit of account, and if AI authors most of that change, then the scarce resource is not "hands that type code" but "eyes that catch regressions, judge model behavior, and own evals." Cycode's ADLC framing captures the shift: "AI agents generate code on their own, pull dependencies, call external tools, and commit changes without a developer ever opening a traditional IDE." Large teams become viable, even necessary, because the review, evaluation, and drift-monitoring surface grows with the number of agent-generated changes.

That is why Gartner's June 2025 prediction (more than 40% of agentic AI projects will be canceled by end of 2027, mostly due to costs, unclear value, and inadequate risk controls) is a hiring thesis, not a warning. The 60% that survive will be the ones who staffed the reliability side heavily from day one.

The scarce hire in ADLC is not the person who ships an agent. It is the person who keeps one honest at 3 a.m.

The ADLC Engineering Director role has near-zero LinkedIn recall

An "ADLC Engineering Director" search on LinkedIn today returns essentially nothing, and that is the opportunity. Anyone whose headline already reflects ADLC vocabulary is self-selecting into a tiny pool of early adopters, exactly the profile a Series A wants leading 5 to 7 teams.

The right target profile has three layers:

1. A staff-plus engineer who shipped a production agent at a real company (not a demo, not a hackathon). Look for postmortems, eval dashboards, and named model versions in their writing.
2. Someone who has managed 20-plus engineers before, ideally through a platform migration. ADLC is a platform migration in disguise.
3. Comfort with the negative signal. They should be able to articulate why "AI coding assistant adoption" is a losing metric and "cost per merged verified change" is a winning one.

That combination sits at the intersection of three shrinking pools. Refolk's index shows LangGraph-listed engineers across the US, UK, Germany, and Sweden total 6,373, heavily concentrated in the Bay Area, London, and Berlin. Filter to people who have also managed teams of 20-plus and shipped an eval suite, and you are looking at a genuinely nameable list.

A concrete playbook for the Stockholm JD (or the next one)

Stop searching headlines, start searching artifacts, and widen geography before you widen seniority. Here is the sequence I would run for a JD like Stockholm's, in order.

• Week 1: build the artifact list. Pull every public MCP server, every LangGraph repo with real usage, and every evals/ directory in agent-related orgs. This becomes your true addressable pool, not the 15 headline matches.
• Week 2: layer in employer signal. Cross-reference against people currently at ProSiebenSat.1, Nokia, and the agent-focused consultancies that top Refolk's LangGraph-plus-Claude-Code list. These are the "already in Europe, already shipping" candidates.
• Week 3: widen to Berlin and London. The 4-country LangGraph pool is 6,373. Sweden is 0.05% of that. Berlin and London together carry the majority. If visa sponsorship is real, so is the relocation pipeline.
• Week 4: reach out on the artifact, not the JD. "Saw your eval harness for tool-call accuracy in [repo]" gets a reply. "I'm reaching out about an exciting opportunity" does not.

The second and third steps are where Refolk earns its keep, because "engineers at Nokia or ProSiebenSat.1 who committed to a LangGraph repo in the last 90 days" is not a query LinkedIn Recruiter can answer, and it is exactly the query that separates the 15-person headline pool from the several-hundred-person real pool.

What this JD signals for the next 12 months

Expect ADLC vocabulary to show up in enterprise JDs within 6 to 12 months, dragged in by the vendors already publishing frameworks (Salesforce Agentforce, IBM, Cognizant, Atlan, Arthur.ai), and expect the sourcing pool to lag the language.

Two practical implications for sourcers and engineering leaders:

• Build your ADLC candidate list now, quietly. The people who match are not job-hunting. They will be poachable in 6 to 9 months when their current agent project stalls into Gartner's 40% cancellation bucket.
• Stop measuring your funnel by LinkedIn InMail response. If 93% of qualified candidates are invisible to headline keyword search, InMail volume is measuring the wrong pool.

The Stockholm JD is the first one to say "SDLC is out, ADLC is in" out loud. It will not be the last. The teams that build a non-LinkedIn playbook before the language spreads will hire the 15, plus the several hundred hiding behind them. Everyone else will fight over the one person the keyword search actually returns.

FAQ

How do I source agentic AI engineers when LinkedIn keyword search returns almost none?

Search for artifacts, not labels. The candidates you want have public evidence in GitHub (evals/ directories, MCP server repos, LangGraph subgraphs with checkpointing, commits to promptfoo or braintrust or inspect-ai) that never makes it into a LinkedIn headline. Tools that read GitHub and the open web alongside LinkedIn will surface roughly 15 times more qualified candidates than headline-only search, based on Refolk's 1-to-15 ratio between headline-searchable and skill-tagged ADLC candidates globally.

What is the Agentic Development Lifecycle (ADLC) and how is it different from SDLC?

ADLC is a software lifecycle designed for systems where LLMs sit at the core of product behavior, with prompts, model choices, and eval thresholds treated as first-class engineering artifacts. IBM's definition explicitly rejects SDLC assumptions, and the agenticlifecycle.ai reference frames the shift as "the unit of account is cost per merged, verified change. Not tokens per developer per month." It is not SDLC plus a coding assistant.

What should an ADLC engineering director role actually screen for?

Screen for shipped production agents with named model versions and measured eval improvements, prior experience managing 20-plus engineers through a platform migration, and the ability to explain why "AI coding assistant adoption" is the wrong metric. Screen out headlines that lead with "Copilot power user" or "prompt engineer." The signal you want is ownership of an eval harness and a drift dashboard, not tool enthusiasm.

Is the 30 to 40 engineer team size realistic for an ADLC-native org?

Yes, and it is likely to become more common, not less. If AI authors most of the code, review and evaluation become the bottleneck, and review scales with the volume of agent-generated changes rather than the number of human authors. That inverts the "small elite team" startup default and makes 30-plus engineer Series A orgs economically rational, provided the reliability side (evals, drift, red-team) is staffed heavily from day one.