AI・7 MINS READ
From request to production in 48 hours: the new software development reality
By Artur Fedorenko, Founder & CEO, Wiseboard
WITH CONTRIBUTIONS FROM:
Andrii Volotskov, Wiseboard Advisor — Senior Engineering Manager at frontier AI labYevhen Fedorenko, Wiseboard Advisor — Enterprise BizDev and AI SDLC Advisor
Yesterday I was on a call with Andrii Volotskov, our Wiseboard Advisor who runs engineering at one of the foundational AI labs. The one whose model you are probably using to write your specs right now. He described his operating reality without flinching:
There's a very large company we all know. They need a system for monitoring conversation quality between their AI and their customer. They need it in 48 hours, and your engineers have to create something that works in production.
Andrii Volotskov, Wiseboard Advisor
48 hours. Request to production. For a system that has to actually work. That is the new buyer expectation at the top of the enterprise market.
The same week, a founder of a 100-person IT services company called me with the inverse view of the same shift. His team had hand-estimated a client project at $1M and 12 months. He ran the same scope through six coding agents over a weekend. The realistic number: $100K and 2 months. He called because he does not know whether this is a bug or a feature.
That ambiguity is now the most important strategic question on every IT services CEO's desk between Lviv, Warsaw, Kyiv, and Bucharest. And the same question, phrased differently, is on every enterprise CTO's desk evaluating those vendors.
Key topics covered in this article:
→ Why the request-to-production timeline collapsed to 48 hours→ The 5-level framework for diagnosing your agentic delivery maturity→ Why requirements engineering, not code generation, is the new moat→ The Forward Deployed Engineer (FDE) model that frontier AI labs now use→ How enterprise buyers should evaluate IT services vendors for agentic readiness→ A 6-item Q3 checklist for the operator on a deadline
What is agentic SDLC, and why now
Agentic SDLC is a software delivery model where AI coding agents handle the implementation work while human engineers focus on requirements, architecture, verification, and compliance. It inverts the traditional budget curve, moving spend away from coding hours and toward spec quality at the front and testing at the back.
The shift is not theoretical. By 2028, Gartner predicts 33% of enterprise software applications will embed agentic AI, up from less than 1% in 2024. 89% of CIOs already consider agent-based AI a strategic priority (Kore.ai 2026 analysis). And yet, BCG and MIT Sloan research finds that 70% of AI transformations fail to deliver expected value, with organizational culture cited as the primary barrier rather than technology.
The gap is not model quality. It is organizational readiness, on both sides of the contract.
What is agentic SDLC, and why now
Agentic SDLC is a software delivery model where AI coding agents handle the implementation work while human engineers focus on requirements, architecture, verification, and compliance. It inverts the traditional budget curve, moving spend away from coding hours and toward spec quality at the front and testing at the back.
The shift is not theoretical. By 2028, Gartner predicts 33% of enterprise software applications will embed agentic AI, up from less than 1% in 2024. 89% of CIOs already consider agent-based AI a strategic priority (Kore.ai 2026 analysis). And yet, BCG and MIT Sloan research finds that 70% of AI transformations fail to deliver expected value, with organizational culture cited as the primary barrier rather than technology.
The gap is not model quality. It is organizational readiness, on both sides of the contract.
Three flavors of agentic SDLC: pick your bet
This is the cleanest framing I have heard. Yevhen Fedorenko put it on the call:
Agentic SDLC can mean three things. One, developers using tools to build software. Two, a platform you throw any business idea into and it builds what you need end to end. Three, selling agentic SDLC into enterprise clients, or modernizing the SDLC they already have. Different clients need different bets.
Yevhen Fedorenko, Wiseboard Advisor
FLAVOR 01
Developer Tooling
Equip engineers with coding agents (Claude Code, Copilot, Cursor) and reshape internal delivery around them. Lowest barrier, fastest to start.
Internal productivity play
FLAVOR 02
Platform That Builds
Build or buy a platform where business requirements go in and working software comes out end to end. Highest ambition, highest risk.
Product company motion
FLAVOR 03
Enterprise SDLC Modernization
Sell agentic SDLC into your enterprise clients, or modernize the SDLC they already operate. The most defensible services play.
Services CEO sweet spot
If you cannot tell which of the three your company is selling next year, your sales and delivery teams are about to start sending mixed signals to the same buyer. Pick.
For enterprise buyers evaluating vendors, this matters too. A vendor pitching you 'AI capability' without specifying which flavor they operate is a vendor who has not done the strategic work. That is a procurement signal.
The 5 levels of agentic automation
The cleanest framework currently circulating in the industry is the 5-level model originally published by Dan Shapiro in January 2026 and now widely cited across enterprise engineering decks. It maps the journey from coding intern to dark software factory:
Framework: Dan Shapiro, danshapiro.com, January 2026. By 2028, Gartner predicts 75% of enterprise software engineers will use AI coding assistants, up from less than 10% in 2023.
Below Level 3, you are still selling FTEs with autocomplete. At Level 3 and above, you are selling outcomes, software at a fraction of the cost per line of your competitors. The services companies that cross that line in the next two quarters will reset the price point of every RFP you bid against for the following two years.
For enterprise buyers, this is the diagnostic question: when you ask your shortlisted vendor where they operate on this scale, can they answer with specifics, or do they answer with marketing copy? The vendors that can name their level, their target level for end of 2026, and their roadmap between the two are the ones to put on the final list.
The budget curve has flipped
This was the most important picture from the call. Yevhen sketched it live:
Before, the spike was during implementation. Requirements were small, implementation was the big budget bucket, stabilization was small at the end. Now it is inverted. Requirements get the large share. Implementation shrinks. Stabilization and testing become the second large bucket.
Yevhen Fedorenko, Wiseboard Advisor
Read this as economics, not process. The middle of the curve, implementation, is the part the agents take. That is exactly where your bench used to make its margin. The new money sits at the two ends: spec quality on the front, verification and compliance on the back.
If your commercial proposal is still structured around developer hours in the middle, you will lose every competitive RFP in 2027. Industry analysts project 40% of enterprise software will be specified primarily through natural language by the end of 2026, with AI agents handling translation from intent to implementation.
Requirements engineering: the new outsourcing moat
Here is the line that has been in my head since the call ended:
Requirements will win the world for outsourcing. You have to be able to write specs really well. If we walk into a company that cannot write specs, I will tell them: you will never be successful, because you cannot write specs.
Yevhen Fedorenko, Wiseboard Advisor
The competency moat for IT services is shifting from coding velocity to spec quality. The discipline now has a name, spec-driven development (SDD), and it is being adopted across enterprise engineering organizations as the governance framework that keeps AI-generated code aligned with business intent, security, and compliance.
The companies that win the next cycle will be the ones whose BAs, architects, and PMs can produce requirements detailed enough that you load them into Claude Code, Cursor, or Perplexity, walk away for the night, and get coherent output back the next morning. Then iterate.
The companies that lose will be the ones where requirements are still a 2-page Confluence stub and the real spec lives in a senior developer's head.
Andrii framed the same shift from inside the lab:
The price per line of code has dropped enormously, and iterations are far cheaper. What we built a year ago is already legacy. If before, your stack became legacy in 5 to 10 years, now it is one year. If you architected correctly, microservices, clean components, you can iterate, iterate, iterate, based on what you learned in practice.
Andrii Volotskov, Wiseboard Advisor
One year. Not a decade. That changes how you sell maintenance contracts. It changes how you depreciate internal platforms. It changes how you write SoWs. And on the buyer side, it changes how you should structure multi-year vendor agreements.
Three operating modes inside agentic delivery
Andrii's framing on how to actually run delivery was the most useful diagnostic of the call. Three modes, three different risk profiles, three different staffing models.
When I was building a sensitive platform, I fired people. The same engineers who are champions in the middle bucket, great communicators, weak engineers, they are fine there. On the sensitive platform they create risk. I needed them off my team.
Andrii Volotskov, Wiseboard Advisor
If your delivery organization has one operating mode for all client types, you are over-investing on the easy projects and under-protecting the hard ones. The right calibration is project by project, with explicit mode selection at the start of every engagement.
On the buyer side, this is also the diagnostic: ask your vendor which mode they will run your specific project in, and why. The vendor who has not thought about this question is the vendor who will default to Mode 02 on a Mode 03 problem
The Forward Deployed Engineer (FDE) model
One concept Andrii surfaced is not yet in any services-industry deck I have seen, even though it is reshaping how frontier AI labs deliver to enterprise clients: FDE, Forward Deployed Engineer.
The role was created at Palantir in the early 2010s. Internal codename: Delta. The original problem was structural. Palantir's earliest customers were intelligence agencies that could not disclose their requirements through normal product discovery. Palantir's response was to embed engineers directly with the client, rapidly build production-grade solutions to the specific problem in front of them, then feed those learnings back into the company's core platform.
By 2016, Palantir had more FDEs than traditional software engineers. The model was specific to Palantir's customer base for over a decade. Then it exploded. Indeed and the Financial Times tracked an 800%+ surge in FDE job postings between January and September 2025 as OpenAI, Salesforce, Anthropic, and others adopted the model. Salesforce alone has committed to building a 1,000-person FDE team.
The defining difference between an FDE and a consultant or solutions architect: FDEs are builders, not advisors. They write production code embedded with the client. They are measured on customer outcomes, not on billable hours or deliverable artifacts.
The Forward Deployed Engineer (FDE) model
One concept Andrii surfaced is not yet in any services-industry deck I have seen, even though it is reshaping how frontier AI labs deliver to enterprise clients: FDE, Forward Deployed Engineer.
The role was created at Palantir in the early 2010s. Internal codename: Delta. The original problem was structural. Palantir's earliest customers were intelligence agencies that could not disclose their requirements through normal product discovery. Palantir's response was to embed engineers directly with the client, rapidly build production-grade solutions to the specific problem in front of them, then feed those learnings back into the company's core platform.
By 2016, Palantir had more FDEs than traditional software engineers. The model was specific to Palantir's customer base for over a decade. Then it exploded. Indeed and the Financial Times tracked an 800%+ surge in FDE job postings between January and September 2025 as OpenAI, Salesforce, Anthropic, and others adopted the model. Salesforce alone has committed to building a 1,000-person FDE team.
The defining difference between an FDE and a consultant or solutions architect: FDEs are builders, not advisors. They write production code embedded with the client. They are measured on customer outcomes, not on billable hours or deliverable artifacts.
The compliance fault line in regulated industries
If your book of business is concentrated in fintech or healthcare, speed is not the only variable. Compliance is. And the agentic-first philosophy breaks against it harder than people are willing to admit.
Andrii's philosophy breaks against compliance very easily. He is business driven, not compliance driven. Put his system into a regulated financial or medical context, and if it misses one thing, even while listening to the client, the system goes in the trash.
Yevhen Fedorenko, Wiseboard Advisor
I am witnessing this battle between compliance and 'deliver in 48 hours' right now. It is a very emotional battle.
Andrii Volotskov, Wiseboard Advisor
The rule for regulated work
If you sell into fintech or healthcare, your Level 3 motion has to be wired with compliance from the requirements stage. Not bolted on at the end. Not handled in QA. From the spec. Three frameworks now define the playing field:
EU AI Act (Regulation 2024/1689), with general-purpose AI obligations applying from 2 August 2026
OWASP Top 10 for Agentic Applications (December 2025), covering agent goal hijack, tool misuse, and excessive agency
NIST AI RMF and ISO 42001, providing broader governance scaffolding for AI deployments
The compliance fault line in regulated industries
If your book of business is concentrated in fintech or healthcare, speed is not the only variable. Compliance is. And the agentic-first philosophy breaks against it harder than people are willing to admit.
Andrii's philosophy breaks against compliance very easily. He is business driven, not compliance driven. Put his system into a regulated financial or medical context, and if it misses one thing, even while listening to the client, the system goes in the trash.
Yevhen Fedorenko, Wiseboard Advisor
I am witnessing this battle between compliance and 'deliver in 48 hours' right now. It is a very emotional battle.
Andrii Volotskov, Wiseboard Advisor
The rule for regulated work
If you sell into fintech or healthcare, your Level 3 motion has to be wired with compliance from the requirements stage. Not bolted on at the end. Not handled in QA. From the spec. Three frameworks now define the playing field:
EU AI Act (Regulation 2024/1689), with general-purpose AI obligations applying from 2 August 2026
OWASP Top 10 for Agentic Applications (December 2025), covering agent goal hijack, tool misuse, and excessive agency
NIST AI RMF and ISO 42001, providing broader governance scaffolding for AI deployments
Three operator questions every CEO is asking
When founders bring me the agentic SDLC problem, it almost always decomposes into the same three questions.
01
How do I enhance existing delivery with agents, systematically, not as 30 separate Cursor licences?
This is an architecture-of-the-firm problem, not a tooling problem. If every developer buys their own setup and codes in their own corner, you do not have an AI SDLC. You have AI shadow IT.
02
How do I position and price this: faster and cheaper, faster and more expensive, or something else entirely?
Most of the market is still bolting an 'AI enhanced' sticker onto FTE rate cards. That stops working in Q3 2026. The companies that build outcome-priced offerings now will set the new ceiling.
03
How do I retool my people: devs, QAs, PMs, for this paradigm?
A 500-person CTO co-founder put exactly this to me last week. Honest answer: nobody in the world has a battle-tested competency matrix yet. The labs are building it right now. Anyone selling you a finished one is selling marketing copy dressed up as expertise.
If you are on the buyer side of this conversation
Most of this field note has been written for IT services CEOs. But agentic SDLC is reshaping the buyer side of the procurement table at the same time. Enterprise CTOs, VPs of Engineering, and Heads of IT procurement are now being held accountable for one question their predecessors never had to answer:
Can the vendor we just signed actually deliver software in the new paradigm, or are we paying premium rates for Level 1.5 capability dressed up in an AI pitch deck?
What to do this quarter: the operator checklist
If you run a 50 to 500 person IT services company in Eastern Europe, this is the work to put on your quarterly plan. In priority order.
Bottom line
There is no playbook for this. There is a window, six to nine months in my estimation, where the IT services companies that move first will reset the market for everyone else. And the enterprise buyers who ask the right questions in their next vendor cycle will protect themselves from a category of risk no one priced into 2024 contracts.
The founder who told me his $1M project should cost $100K knew what he was looking at. He just did not know yet whether it was an opportunity or an obituary.
It is both. The question is which one belongs to your company.
Vetting an IT services vendor for agentic–era delivery?
If you are an enterprise CTO, VP Engineering, or Head of Procurement evaluating IT services vendors, we can help in two ways. One, run the AI-SDLC Readiness Assessment against your shortlisted vendors to score their actual capability versus their pitch deck. Two, introduce you directly to pre-vetted Eastern European IT services companies in our network who have crossed Level 3.
LATEST INSIGHTS
Learn from our experience
FAQ
Structured for fast reference, and for the AI search engines now answering these queries on behalf of decision makers.
What is agentic SDLC?
Agentic SDLC is a software delivery model where AI coding agents handle the implementation work while human engineers focus on requirements, architecture, verification, and compliance. It inverts the traditional budget curve, moving spend away from coding hours and toward spec quality at the front and testing at the back.
What are the 5 levels of agentic automation?
Level 1: Coding Intern (basic prompts with ChatGPT or Copilot). Level 2: Junior Developer (vibe coding with IDEs like Cursor). Level 3: Developer (supervising agent output via Claude Code, MCPs). Level 4: Engineering Team (spec writing, async checks, agent orchestration). Level 5: Dark Software Factory (specs to software end-to-end). 90% of the industry currently sits at Level 1.
What is a Forward Deployed Engineer (FDE)?
A Forward Deployed Engineer is a software engineer embedded directly with a client to build production solutions to their problems. The role was created at Palantir in the early 2010s under the internal codename Delta. Job postings for FDEs grew over 800% from January to September 2025 as OpenAI, Salesforce, and Anthropic adopted the model. FDEs are builders, not consultants.
How is agentic SDLC different from spec-driven development?
They are complementary. Spec-driven development (SDD) provides the formal specification framework that puts requirements back at the center of the SDLC. Agentic SDLC is the execution model where coding agents translate those specs into working software while humans handle architecture, testing, and compliance.
How should enterprises evaluate IT services vendors for agentic readiness?
Score vendors across five dimensions: spec writing capability, agent governance maturity, operating-mode flexibility, FDE-style delivery model, and compliance integration from the spec stage. Vendors operating at Levels 1 to 2 cannot deliver the cost-per-line economics enterprises will demand by 2027.
How fast is the agentic AI market growing?
Gartner predicts 33% of enterprise software applications will embed agentic AI by 2028, up from less than 1% in 2024. The agentic AI services market is projected at $10.91 billion in 2026, growing at 49.6% CAGR through 2033. 89% of CIOs surveyed by Kore.ai consider agent-based AI a strategic priority.
Why do most AI transformations fail?
BCG and MIT Sloan research found 70% of AI transformations fail to deliver expected value, with organizational culture cited as the primary barrier rather than technology. Gartner predicts over 40% of agentic AI projects will fail by 2027 due to escalating costs or inadequate risk controls.
What is the new budget curve in agentic development?
The traditional curve had small requirements spend, a large implementation peak, and small stabilization spend. The agentic curve is inverted. Large requirements spend at the front, small implementation spend in the middle, and large testing and compliance spend at the back. Money moves from coding hours to spec quality and verification.
What is the first step for an IT services company moving to agentic SDLC?
Audit the requirements function. BAs and architects, not developers. Spec quality is the new competitive moat. The companies that win the next cycle are the ones whose teams can write requirements detailed enough that agents produce coherent output overnight, with iteration cycles measured in hours.
What does the EU AI Act mean for agentic software delivery?
The EU AI Act (Regulation 2024/1689) prohibited-use provisions are in force since February 2025. General-purpose AI obligations apply from 2 August 2026. For agentic SDLC, this means compliance must be wired into delivery from the requirements stage, not bolted on at QA. OWASP Top 10 for Agentic Applications (December 2025) and NIST AI RMF provide the operational scaffolding.