DEV Community: HR Pulsar

Why AI Agents Have Become a Core Feature: Analyzing the Pricing Strategies of Lattice and Its Competitors

HR Pulsar — Tue, 16 Jun 2026 15:26:56 +0000

2025 - the Moment Everything Changed
An HR director opens the Lattice dashboard. They notice the AI-powered functionality and think, "Okay, that's a nice feature."
Then they see the invoice. Then they put their face in their hands.

Six months later, that same HR director looks at a competitor whose AI features are included in the base plan by default. No extra fees. No separate "AI Package." It's just there.

That's not a coincidence. It's a signal that a paradigm shift has taken place.

What Happened: AI Stopped Being Optional

A few years ago, AI in HR software was positioned as a premium feature.
A luxury add-on. Something like leather seats in a car—nice to have, but not essential.
Companies paid $50–100 per employee per year for core functionality (assessments, PDPs, organizational structures) and viewed AI as an exotic extra.

Lattice was one of the first major players to recognize where things were headed.

In 2024–2025, Lattice made a move that only looks rational if you understand what comes next: They embedded AI directly into their entry-level plan.
Not as a separate "AI Premium" package. Not as an add-on. Into the plan that everyone pays for. At first glance, it looked irrational. Why reduce average revenue per user when you could sell AI as an additional module?

Because the base plan stops being attractive without AI.

Why It Happened So Quickly: The Vocational School Effect

By 2026, every employee uses between three and eight AI tools as part of their daily work.
This isn't a hypothesis. It's an observable reality: ChatGPT. Claude. Perplexity. GitHub Copilot. Specialized internal bots. People have become accustomed to AI being part of the background fabric of work itself.

Now imagine an employee opening a talent management system that feels like a web application from 2015. No AI. No assistant. No personalized recommendations.

When someone has Claude in one tab and your system offers checkboxes in a form, that's not a contrast. It's incompatibility.

Lattice understood this. If they didn't integrate AI into the core product, customers would eventually move to someone who did.
This wasn't a marketing decision. It was a survival decision for the category.

Pricing Strategy: How This Changes ROI

Let's talk about the economics.
The Old Model (2020–2023)

Base Tier: $50/user/year
└─ Оценки, PDP, org structure

Premium Tier: $100/user/year
└─ + Аналитика, интеграции

AI Add-on: +$30/user/year (опция)
└─ Рекомендации по развитию, автоматизация ревью

The problem: Only 10–15% of customers paid for the AI add-on.
Most customers chose the base plan and stayed there.

The New Model (2025–2026)

Base Tier: $65/user/year
└─ Assessments, PDPs, organizational structure,
   AI assistant included by default

Premium Tier: $130/user/year
└─ + Advanced analytics,
   Custom LLM,
   Enterprise integrations

At first glance, ARPU appears to decline in the base tier. It looks like a strategic mistake.

In reality, it was cross-subsidization.

Here's what actually happened:
1. Reduced Churn Customers who previously viewed AI as too expensive no longer needed to make a separate purchasing decision. Retention improved by 40–50%.

2. A Clear Upgrade Path Customers experienced AI in the base plan and started asking: "What could we do with even better recommendations?". Conversion from Base to Premium increased by 2–3x.

3. The Data Network Effect When everyone uses AI, more data is generated. More data improves machine learning models. Better models produce better recommendations. Better recommendations improve the product. It's a self-reinforcing cycle.

4. A Market Signal The message is straightforward: "Without AI, you're no longer competitive." That puts pressure on Workday, Personio, and everyone else in the category.

Real ROI for Customers
Previously:

The system helped users fill out assessment forms.
PDPs were created manually by managers (4–6 hours per employee annually).
Leveling and grading required HR meetings (40–80 hours annually per team).

Now, with AI included in the base plan:

AI generates assessment drafts (30 minutes instead of 2 hours).
AI recommends PDP items based on competencies (saving roughly 3 hours per employee).
AI suggests grade levels with supporting rationale (reducing alignment time by about 50%).

For a team of 100 employees, that's 250–300 hours saved per year. At an HR labor cost of $50–100 per hour, that's $12.5K–30K annually. Moving from $50 to $65 per user costs approximately $1.5K per year.

The payback period is measured in weeks.

How Competitors Are Responding: Patterns in 2026

No one can afford to ignore this trend. Here's what's happening across the market.

Lattice

AI is integrated everywhere: assessments, development recommendations, attrition prediction, talent discovery.
Pricing remained stable while premium tiers expanded with cloud functionality and integrations.
_Strategy: _dominate the mid-market segment (500–2,000 employees) through consistently high baseline quality.

15Five / Leapsome

Started following the same approach by including AI in their base plans.
Reduced prices by 15–20% as a defensive move.
Challenge: they lack the data volume available to Lattice for model training.
Positioning: "similar to Lattice, but cheaper"—effective in price-sensitive segments.

Workday / SAP SuccessFactors

Attempted to add AI, but their underlying architecture works against them.
Their core systems are built around jobs and organizational hierarchies rather than competencies.
Adding AI on top feels like putting a bandage on a tank.
Market perception: increasingly viewed as late-moving incumbents.

Eightfold / Gloat

Positioned themselves as AI-first platforms from the beginning.
But their AI offerings primarily target Fortune 500 organizations.
Smaller companies compare them to platforms like Lattice that work effectively for organizations with as few as 50 employees.
Challenge: expensive and difficult to scale downmarket.

Personio, Zelt, and Other European Vendors

Adoption has moved more slowly due to the regulatory environment, particularly the EU AI Act.
But they recognize the same reality: failing to integrate AI means falling behind.
Approach: localize AI capabilities, ensure compliance, then scale.

Why This Trend Is Irreversible

There are three reasons why embedding AI into entry-level plans isn't a temporary strategy. It's the new baseline.

1. Customer Expectations Have Permanently Shifted
Every day, employees:

Write emails with AI assistance.
Debug code with AI assistance.
Generate images with AI assistance.

Then they open a performance management platform, fill out forms manually, and are asked to pay extra for AI-assisted development planning.

That feels absurd. The expectation shift is permanent.
No company can retrain users to stop expecting AI. Trying to move AI back into an optional add-on would damage credibility.

2. Data Has Become a Competitive Asset
Previously: More users → more server load → higher costs.
Today: More users → more training data → better models → better recommendations.

By embedding AI into its base plan, Lattice effectively made a long-term bet on data. ARPU may have changed, but training data grew exponentially.
Organizations that fail to recognize this lose speed-to-insight.

3. Compute Costs Have Fallen
In 2020, processing a single assessment through an LLM could cost $0.10–0.50. Today, it's closer to $0.01–0.05.

This is similar to what happened with cloud infrastructure in the late 2000s. Once cloud became inexpensive enough, maintaining on-premise servers stopped making economic sense.

The same thing is happening with AI.

What This Means for Companies Evaluating HR Platforms in 2026

If you're evaluating HR software today, pay attention to the following signals.

Red Flags
❌ AI is sold as a separate add-on.
❌ The "AI assistant" is merely a chatbot that waits for user prompts.
❌ AI is advertised without specific use cases.
❌ The company still markets AI as a luxury feature.

Green Flags
✅ AI is embedded directly into core workflows (assessments, PDPs, recommendations).
✅ No additional payment is required for basic AI usage.
✅ Time savings are clearly quantified.
✅ The vendor openly discusses limitations and keeps humans responsible for final decisions.

Where HR Pulsar Fits Into This Landscape

We recognized this shift before the market officially acknowledged it.

Our Position
We're building a platform designed around competencies rather than job titles. That's an architectural choice. And it makes AI integration feel natural rather than bolted on.

Why does that matter?
Lattice integrated AI into an architecture built around jobs and organizational structures.
It works. But it's a bit like adding a passenger seat to a cargo truck—functional, but not elegant.

We designed a competency graph from the start. Within that model, AI helps:
✅ Match competencies (pgvector embeddings against 4,800+ role models)
✅ Recommend development paths (which competencies should be developed for a target role)
✅ Suggest career moves (internal talent marketplace: employee + AI toolkit → project)
✅ Manage hybrid teams (people and AI agents visible within a single system)

And this is not a premium feature. _ It's foundational._

We also made a decision that would be difficult for Lattice to make without rebuilding its core architecture: We embedded an enterprise AI tools registry. The Workforce Map doesn't just show people. It shows which AI tools are approved, who owns them, and where risk exists.

This is more than workforce management. It's visibility into the reality of hybrid teams, something no other platform currently provides.

The Brutal Math: What Happens Next

Let's extrapolate.

2026–2027 (Current Phase)

Every top-10 HR platform will include baseline AI functionality.
ARPU will decline by 20–30%.
Retention will increase by 35–50%.
Startups that fail to integrate AI will lose rankings and relevance.

2027–2028 (The Next Wave)

AI systems will require specialized operators and configuration experts.
New categories will emerge around AI accountability, governance, and auditability.
The EU AI Act and NIST AI RMF will become operational requirements rather than optional frameworks.
Companies will discover they're using dozens of AI tools that nobody tracks.

2028 and Beyond

AI will become as invisible and as essential to HR software as SQL is to backend systems.
The question will no longer be whether AI exists.
The question will be how effectively it operates within your architecture.
Companies that layered AI onto legacy systems will struggle.
Competency-centric platforms with built-in hybrid workforce management will gain structural advantages.

For Builders: Why This Matters
If you're developing HR software today, here's what you need to understand.

1. Architecture Is Destiny
If your platform is fundamentally built around job titles, integrating AI will be painful. Start with a competency graph.

2. LLM Errors Are a Feature, Not a Bug
Treat AI output as a recommendation, not a decision. Humans must retain final authority. This reduces legal exposure and increases trust.

3. Data Flows Matter
Where does the AI get its data? From a single organization? Or from an anonymized population? The answer determines which features are possible. Self-hosted systems can never provide industry-wide benchmarking. That's not a design limitation. It's mathematics.

4. Open Source Signals Trust
Technical buyers evaluate open-source AI systems differently. Not as black boxes.

As tools they can inspect, understand, and extend.

# Пример: как мы думаем о встраивании AI в HRPulsar

class CompetencyMatcher:
    """
    AI Fluency — это не магия. Это pgvector embedding.
    Сотрудник имеет компетенции. Роль требует компетенции.
    Косинусное расстояние между векторами = match score.
    """

    def recommend_development(self, employee_id, target_role_id):
        # LLM помогает сформулировать смысл рекомендации
        # Но мэтчинг — это чистая математика
        employee_embedding = self.get_embeddings(employee_id)
        role_embedding = self.get_embeddings(target_role_id)

        gaps = role_embedding - employee_embedding
        # ^ Это — список компетенций для развития

        recommendation = self.llm.generate(
            prompt=f"Сотрудник {name} хочет перейти на роль {role}. "
                   f"Ему нужно развить: {gaps}. "
                   f"Напиши конкретный PDP."
        )

        # ВАЖНО: это — предложение, не приказ
        return {
            "recommendation": recommendation,
            "confidence": self.calculate_confidence(gaps),
            "final_decision_owner": "manager"  # Не LLM!
        }

An Honest Assessment: What Could Go Wrong

Embedding AI into the base plan isn't risk-free.

**Risk #1: **Margin Erosion Without Scale
If you've integrated AI but only have 50 customers, revenue decreases while LLM infrastructure costs increase.
You're underwater.

Mitigation: Integrate AI once you reach meaningful scale (roughly 200–500 customers with 50+ employees each). Until then, keep it optional or in alpha.

Risk #2: Poor AI Can Destroy Trust
If your LLM generates invalid leveling recommendations and a manager blindly follows them, you've created a potential legal problem.

Mitigation: Be transparent about limitations.
"AI recommendation" does not mean "final decision." Educate customers. Log everything.

Risk #3: Regulation Moves Faster Than You Do
The EU AI Act requires documentation for high-risk AI systems. The NIST AI Risk Management Framework is increasingly becoming the de facto standard. Ignoring compliance while integrating AI creates future pain.

Mitigation: Start compliance work now, in parallel with AI development.

Closing: This Isn't the End of the Story

AI in the base tier of HR software isn't the peak of the wave. It's preparation for the next phase.

Because within the next 18 months, another shift will occur: M*anaging hybrid teams (people + AI agents) will become a core capability rather than a peripheral feature.*

Companies that currently use AI merely as an assistant—"here's a development recommendation"—are only preparing for the transition.

When organizations begin working alongside real AI agents that actively perform work, workforce management systems will need to see and manage those agents the same way they manage people.

That's when the real competition begins.

Until then, if you're building or selecting an HR platform, remember: AI in the base plan is no longer a differentiator. It's the cost of entry.

Everything else comes down to execution quality and architectural decisions that either enable—or prevent—the next stage of evolution.

Interested in how we're building this inside HR Pulsar?
Open source. 228 endpoints. Fully typed OpenAPI.
Spin up Docker, import your employees, launch an assessment, and see for yourself.
If you don't like something, open an issue.

Use it completely free https://app.hrpulsar.com/

Why we calibrate the indicator Total, not the raw scores

HR Pulsar — Tue, 02 Jun 2026 11:35:12 +0000

A 360 review collects answers from three to ten people about one person. Then the reviewer sits down to "calibrate" the result. The industry default is one of two paths: rewrite individual respondent scores, or override the per-competence percent at the very end. We tried both. Neither aged well. This week we shipped a third option, and we're not done thinking about it.

This post is half a release note, half an open question.

The raw-score calibration trap

The natural place to put calibration is on the raw answer. Respondent A gave 3, B gave 5. The reviewer thinks the real answer is 4, so they edit one of those numbers and recompute.

Two problems with that.

First, you've now lied about respondent A. Anyone who reopens the survey sees a number A never gave. The anonymity guarantee — the thing you spent ages explaining to the org — becomes "anonymous except when the reviewer disagrees."

Second, the indicator average is the wrong place to override anyway. A competence has five to ten indicators across three skill levels. The reviewer's gut isn't "the average of these answers should be 4.2." It's "this indicator at this level: the person clears it." Different math.

The other common shape is to leave raw answers alone and override the per-competence percent at the bottom. Cleaner. But you've now hidden which indicators the reviewer disagreed with. The output is correct, the audit trail is gone.

What we built

Reviewers pin a Total per indicator. Not on the raw answers. Not on the competence percent. On the indicator itself.

python
class CalibratedIndicatorTotal(Base):
    __tablename__ = "assessment_calibrated_totals"
    assessment_id: Mapped[UUID] = mapped_column(ForeignKey("assessments.id"))
    indicator_id: Mapped[UUID] = mapped_column(ForeignKey("competence_indicators.id"))
    skill_level_id: Mapped[UUID] = mapped_column(ForeignKey("skill_levels.id"))
    total: Mapped[int]  # 0..100
    calibrated_by: Mapped[UUID] = mapped_column(ForeignKey("users.id"))
    calibrated_at: Mapped[datetime]

Raw answers stay intact and stay visible in the same UI. The per-competence percent inherits the override — anywhere a calibrated Total exists, a calibrated chip surfaces in the breakdown so the audit trail is one click away.

Two operational rules around it:

Submissions lock while calibration is in progress. Late respondents can't trickle answers into a result the reviewer is already pinning. The assessment status flips to a calibrating sub-state and the take-assessment route returns 409.
Cancel calibration wipes every Total and restores raw averages. Not a soft delete — clean wipe. Cancel and you're back to the view a respondent saw.

The math: if a competence has indicators I1..I4 at skill level Intermediate and the reviewer pins Totals for I1 and I3 only, the competence percent at Intermediate is mean(pinned(I1), raw_mean(I2), pinned(I3), raw_mean(I4)). We don't silently propagate one pinned indicator to a whole level.

What we don't know yet

Three places where we made a call and aren't sure it survives contact with real teams.

Should the lock be hard or soft? Right now we hard-lock the whole assessment during calibration. The argument for soft is operations — reviewers don't need to chase the last respondent before they can start. The argument for hard is data integrity — calibration math is unstable when its inputs are still moving. We picked hard. Open question whether that's right for orgs with long review windows.

Should calibration carry a written rationale? The data model has room. The UI doesn't surface a comment field yet, because the moment you add a "why did you override this" field, reviewers either skip it or write "see meeting notes." Neither helps the person being reviewed. So: what would actually be useful here — a free-text rationale that nobody fills in, or a structured tag (e.g. "respondent bias", "context the survey missed") that constrains the input?

Re-entry semantics after Cancel. If a respondent submits after calibration was cancelled, their answer re-enters the raw average automatically. The alternative was an explicit "include late submissions" toggle. We picked auto-include because Cancel already feels like a reset button. But that's a guess.

Four years ago we started building an HR platform

HR Pulsar — Mon, 18 May 2026 10:56:41 +0000

Not because HR-tech looked exciting. Mostly because every company we talked to had the same setup:

grades in Excel
360 reviews in Google Forms
competency matrices buried in Confluence
“performance systems” that nobody wanted to open twice

And eventually: “Let’s just build our own thing.”
So we did.

Then this year we made a bigger decision: we rewrote the platform from the ground up — and open sourced it.

Meet HRPulsar

An open source talent management platform for teams where every employee already works with AI tools.

Most HR systems still model companies like it’s 2015: employee → position → annual review.

But that model is breaking. Roles change every quarter. Skills decay faster than job titles. Half the team uses AI agents nobody tracks. And “career frameworks” are often PDFs with better branding.

We built HRPulsar around a different assumption:
Skills are the stable unit. Not positions.

So the platform is competency-first at the core: competency graphs, grade systems tied to behavioural anchors, 360 assessments, internal talent marketplace, development plans, AI fluency tracking, AI workforce registry
Yes, AI workforce registry.

Because “Shadow AI” is becoming a real operational problem: teams use 15 different AI tools, security doesn’t know about half of them, and HR systems pretend none of this exists.

So we added:

AI tool registry
employee ↔ AI tool assignments
oversight levels
workflow ownership
audit trails
workforce maps for hybrid teams (human + AI)

Useful for compliance & security. But mostly because companies should probably know which AI systems are already part of their workforce.

Technically, HRPulsar is

FastAPI
Next.js
PostgreSQL
Redis
200+ API endpoints
~1000 automated tests
Docker-first deployment
AGPLv3

No “book a demo”. No sales call before seeing the product.
And one thing we decided early: HRPulsar will stay free for individuals and small teams.

Every workspace gets a monthly pool of renewable credits. Enough to run the core platform without turning “try the product” into a budgeting discussion.

And yes — self-hosting is fully supported. Because HR data is sensitive.
And “trust us with your entire workforce structure” is a pretty big ask from a black-box SaaS vendor.

We’re preparing the public GitHub release right now.
*The repository goes live at the end of May. If you want early access, release updates, or just want to watch the project evolve — join the waitlist.

LLMs are useful ONLY for competency matching and draft recommendations

They are not qualified to decide someone’s promotion. We’re very explicit about that in the product.

LLMs are useful for competency matching and draft recommendations.
They are not qualified to decide someone’s promotion. We’re very explicit about that in the product.

Current limitations?
We don’t cover recruiting yet. ATS/recruiting is on the roadmap, but not today.

Would genuinely love feedback from developers building internal tools, people running self-hosted infrastructure, HR engineers, teams experimenting with AI-heavy workflows... and especially from anyone who has ever tried to manage competencies in Excel. We know you exist.

try it on website

Why AI adoption fails inside companies

HR Pulsar — Thu, 07 May 2026 12:15:48 +0000

A company buys ChatGPT Enterprise. Then:

marketing uses it for copy
one engineer automates half their workflow
another refuses to touch it
managers say “we should use AI more”
nobody knows what “more” means

Three months later, leadership asks the inevitable question: “So… are we actually getting ROI from this?”. Silence. Not because AI failed.
Because adoption inside companies is mostly random. And random systems don't scale.

Most companies approach AI rollout like this:

Buy tools
Announce initiative
Hope employees figure it out

That works for maybe two weeks. After that, you get what every company gets:

inconsistent usage
inconsistent output
no shared standards
no visibility into who’s actually effective with AI

Useful? Sometimes.
Measurable? Not really.

The uncomfortable part: most companies still evaluate people like AI doesn’t exist.
Performance reviews ask communication, ownership & collaboration.
Fine. But now we also need to ask:

can this person delegate effectively to AI?
can they verify AI output?
do they know when not to use it?
are they faster because of AI — or just noisier?

Because there’s a difference between “uses AI” and “works effectively with AI” is enormous.
And this is where most AI adoption projects quietly break. Not on infrastructure. Not on tooling. On management.
No competency model. No measurement system. No shared definition of “good AI usage” - just licenses and optimism.

The weird part is that companies already solved this problem once.
We don’t say: “Everyone has Excel now, good luck”. We train people, define expectations, measure proficiency.
AI will end up the same way. Except the impact is bigger.

At HRPulsar, we’ve been thinking about this a lot. Not as “AI replacing employees”. But as: how do you systematically measure and develop AI fluency inside teams?
And honestly, we don’t think the industry has good answers yet.

Especially for:

role-specific AI competencies
measuring real usage quality
separating employee contribution from AI contribution One thing already seems obvious: buying AI tools is easy, building an organization that actually knows how to use them is the hard part.

Curious how other teams are handling this right now. Do you actually measure AI adoption in any meaningful way — or is it still mostly vibes?