Sridhar Sambangi

H‑1B Fees & U.S. Tech Jobs: Time to Reset the Incentives President Trump just imposed a $100,000 application fee for new H‑1B visas. Whether you cheer or jeer, it’s a wake‑up call: we’ve built an industry around labor arbitrage that sidelines American grads while mega‑cap tech posts record profits. Here’s the reality behind the rhetoric: Who benefits most from H‑1B today? In FY2025, ~71% of H‑1B approvals went to Indian nationals, and the biggest users include both Big Tech and large IT outsourcers. That’s a concentration risk that any balanced system should address. The pipeline is bigger than just H‑1B. The U.S. also authorizes large numbers of OPT/STEM OPT/CPT work permissions for foreign students each year (~194,554 OPT, 95,384 STEM OPT, 130,586 CPT in 2024), which many firms treat as a feeder into H‑1B. Follow the money. The top five Indian IT firms generate $44B+ per year from U.S. clients, illustrating how deeply U.S. corporate IT depends on this low‑cost model. And the cash flows don’t stop here. India received a record $135.46B in remittances in FY25; advanced economies—led by the U.S.—supplied ~27.7% of the prior year’s inflows (roughly $37–$38B from the U.S.). That’s a powerful financial tailwind for offshoring models. Bottom line: The “skills shortage” explanation doesn’t fit the full picture. The incentives have rewarded lower-cost placements over building American talent—even as firms report historic margins. What I support (and why): Target high‑end roles, not cheap pipelines. Keep the new fee focused on new H‑1Bs and pair it with strict wage floors and enforcement so visas go to genuinely hard‑to‑fill, top‑quartile roles—not lowest‑bid body‑shopping. Close the outsourcing loopholes. Require direct employment, ban “benching,” and restrict third‑party client‑site placements that turn visas into staffing channels. Diversity guardrails, not dominance. Establish global talent quotas so no single nationality can dominate annual H‑1B allocations, while keeping merit‑based selection intact. Build American talent now. Tie access to visas to measurable U.S. training/apprenticeship investments, paid internships for domestic CS majors, and partnerships with community colleges. Pro‑diversity, pro‑merit, pro‑America. This is about fixing incentives and rebuilding the career ladders here at home—not disparaging individual immigrants who play by the rules. ❤️ On a personal note: My son is studying computer science. He pours his heart into it—and we invest our hard‑earned savings in his education. He shouldn’t have to compete with artificially underpriced labor because a loophole lets companies sidestep fair market pay. We must build a system where young Americans like my son can build decent, dignified careers in the country they call home. Anything less is unacceptable. And if a tech company disagrees, then it’s not acting in America’s interest.

2

2 Comments

Most vibe coders are optimizing the wrong layer. There’s a post going around showing how a “non-technical” prompt and a “senior engineer” prompt produce wildly different results for the same task. The takeaway people grabbed: write longer, more detailed prompts. That’s not the lesson. Here’s what usually happens. A deal team needs to review 6,000 documents in a virtual data room for a mid-market acquisition. Someone asks an AI to flag every contract with a change-of-control provision. They get a script that searches for the phrase, pulls matches, and dumps them into a spreadsheet. Half the results are boilerplate. Key risks in assignment clauses and consent requirements get missed entirely. Nobody knows what the AI skipped or why. An experienced builder doesn’t write a better search query. They ask different questions first. What document types are actually in this data room? Are we looking for explicit change-of-control language, or any provision that could block or complicate closing? Do amendments override the base agreement? Are side letters in the set? What’s the priority: speed to first-pass flags, or completeness for the disclosure schedule? What happens when a clause is ambiguous and needs deal-team judgment? That reframes the problem before a single line of code gets written. The gap between a weak AI output and strong output isn’t verbosity or technical vocabulary. It’s the quality of the constraints. Success criteria. Tradeoff decisions. Knowing what not to build. Vibe coding gets dramatically better when you stop describing the task and start defining the boundaries around it. I Try this: next time you prompt an AI to work through a diligence set, write three sentences before the request. What’s a real risk versus noise? What should get escalated instead of auto-flagged? What’s the consequence of a false negative? That alone will change the output more than any prompt template. Subscribe to PossibLaw. Substack. Podcast. Tools to become a builder. We run custom training for legal teams who are done watching from the sidelines. Let’s talk.

2

2 Comments

Read the new 300 plus page survey on code LLMs and agents so you do not have to. It is about how tools like Copilot, Cursor, Claude Code and others turn natural language into working code and reach over 95 percent success on standard benchmarks. Here is what actually matters for your team. 1. Make AI assistance the default, not the bonus The paper is clear. We already moved from manual coding to AI assisted and now to AI driven and even AI autonomous in the roadmap. If your engineers use Copilot only when they feel lazy, you lose. Set one standard workflow for everyone. Examples. Pair programming with a code model on every feature. Using natural language for small refactors. Let the model write the first draft and the human polish it. You get less boring work, fewer context switches and faster feature cycles. The good teams already treat code LLMs as part of the IDE, not as a toy. 2. Give the model full context of your system The survey keeps repeating one pain point. Real world projects fail when the model does not see enough of the codebase and the docs around it. So you need a proper code intelligence layer for the team. Search over the monorepo. Embeddings for services and APIs. Links to tickets, design docs, logs. Retrieval on top of that. Then the assistant can answer things like What is the real entry point for payments Where do we validate KYC How do we deprecate this feature safely Your people stop guessing and stop breaking hidden contracts between services. 3. Standard prompts plus serious evaluation The paper goes deep into prompting patterns like chain of thought and multi step planning, and then into new ways to evaluate code besides simple pass or fail. For a company this turns into two moves. First, create a small library of prompts for your stack. Example. Write unit tests in our usual style Refactor this into small functions and keep names stable Explain this file to a junior engineer Second, judge the output for more than correctness. Check clarity, performance, security, and how easy it is to maintain, just like the survey suggests for modern benchmarks. You get consistent output and you can finally trust what the tool gives your team. 4. Use agents for the boring multi step work One of the strong parts of the survey is agents. Systems that decompose tasks, call tools, fetch docs and move through several steps without a human holding their hand. This is where teams win big. Documentation generation. API client updates across many services. Mass refactors with tests. Project setup and scaffolding for new microservices. You do not want your best engineers to click through thirty files for the same pattern. You want one agent to do it and a human to review the final merge request.

3

3 Comments

H-1B Lottery Myth: “Level 4 = 4 Entries” There’s growing confusion in the market that filing an H-1B at Wage Level 4 gives “four lottery entries,” Level 3 gives three, Level 2 gives two, and so on. That is not how the current USCIS cap lottery works. Under the active beneficiary-centric selection system, the lottery is based on the unique beneficiary (passport number) — not wage level, not salary amount, and not work location. The rule is simple: One beneficiary + one registration = one lottery entry. If an employer registers four different employees — whether they are: • Level 4 in New York • Level 4 in North Dakota • Level 3 in New York • Level 2 in Texas Each employee receives one entry in the lottery. Total registrations: 4 Total lottery entries: 4 Not 4 + 4 + 3 + 2. The confusion likely comes from earlier discussions about a theoretical wage-weighted selection model. However, under the currently implemented USCIS process, wage levels do not multiply lottery chances. Wage level may affect petition strength after selection — including role credibility, prevailing wage compliance, and documentation quality — but it does not change lottery probability. Understanding this distinction is critical for employers planning cap strategy and setting realistic expectations. Disclaimer: This post is for general informational purposes only and does not constitute legal advice. Employers and beneficiaries should consult qualified immigration counsel for guidance specific to their individual circumstances. — I V Reddy

4

Technology companies can scale software quickly. Contract frameworks often don’t scale with them. I see this often in healthcare technology environments where EMR integrations, SaaS deployments, and vendor ecosystems expand faster than the contracts governing them. When contract language is disconnected from technical execution, problems show up quickly: • unclear responsibility for data transmission and interface builds • service levels that don’t reflect real operational workflows • cybersecurity obligations misaligned with system architecture • repeated contract revisions across every new customer agreement AI tools can review contracts faster. But AI cannot interpret how complex healthcare technology systems actually operate across integrations, vendors, and regulated data environments. The gap between legal language and technical execution is where many negotiations begin to stall. Over the years I developed a proprietary analytical framework the Operational Contract Alignment Method™ designed to identify where contract terms fall out of sync with operational reality before those gaps turn into negotiation bottlenecks or downstream implementation risk. Acceptance Paralegal Services LLC operates at the intersection of legal contracting and healthcare technology infrastructure across EMR, EHR, EDI, and SaaS environments. When contract language reflects how systems actually function, organizations move faster and reduce operational risk across their vendor ecosystems. #HealthTech #SaaSContracts #HealthcareIT #ContractStrategy #AIinBusiness

Why do EMR implementations take 9 months? Because the system requires humans to learn its complexity. Traditional EMR implementations drag on for 6-9 months. Sometimes longer. Agencies delay go-lives, extend training windows, run parallel systems. Staff spend weeks learning where to click, what fields to fill, which workarounds to use. We're doing implementations in 60 days. Not because we cut corners. Because the system is designed differently. When AI handles documentation, coding, compliance, and billing automatically, there's less for staff to learn. When workflows adapt to how clinicians actually work, there's less to configure. When the system does the heavy lifting, training becomes orientation, not education. Implementation timelines reveal something important about a system. If it takes 9 months to learn, the complexity lives in the wrong place. Your technology should learn your agency. Not the other way around.

10

Why Riverside CA – the #18 US Construction Market – Commands 10% of California's Spend and Projects $23.8B by 2030. At BiltData.ai, we have scrutinized over 100 million construction equipment rental and dealer transactions to deliver predictive buyer intelligence that reveals to OEMs, Dealers and Rental Companies exactly what customers (aka buyers) spend on rentals, owned fleet and product support Riverside-San Bernardino-Ontario is not just another construction market—it is a powerhouse accounting for 10% of California's construction activity in 2030, ranking #16 in data centers, and poised for explosive growth. This is not guesswork; it's a clear edge in a market with $23.8B in 2030 construction spending: residential leads with $9.52B, followed by commercial at $7.14B, industrial at $4.76B and infrastructure at $2.38B. The real question: Why settle for flying blind? BiltData's Lead(er) tool combines construction industry expertise with predictive intelligence for rentals, fleet, parts & service; delivering a map-driven tool that’s simple, serious, and unmatched. See how 'completing the picture' can give your business an edge to sense demand, know triggers, know more than the competition and decide with confidence—every time. Your best quarter is 3 clicks away.

7

1 Comment