TechSoft

Whitepaper · Public

AI-Assisted Development: The TS-2VE (Verified-Enterprise) Framework

How Anthropic's IPO validates a new era. How ABB proves it works. How to adopt it safely.

Public review version. Share feedback and questions using the comment section at the bottom. All hours are internal estimates for a product the size of ABB, not measured actuals. No external sources were used. Version 2.0 · November 2026.

Why now

Anthropic (maker of Claude) went public on November 19, 2026, raising $81.3 billion in the largest IPO on record. This validates AI-assisted development as a mainstream practice. The question is no longer whether to adopt it, but how to do it safely at scale. TechSoft has an answer.

The TS-2VE Framework: Speed with safety

The TS-2VE (Verified-Enterprise) Framework is a methodology for building production systems with AI tools. It sits between traditional developer-typed code (TS-1HC) and autonomous agents (TS-3SL/TS-3ML). It is proven, repeatable, and safe.

  1. The opportunity

    Anthropic's IPO (November 2026) validates AI-assisted development as a mainstream practice. The question is no longer whether to adopt it, but how to do it safely at scale.

  2. The framework

    TS-2VE (Verified-Enterprise)—AI coding with engineering discipline—delivers 15–35 owner hours for a product the size of ABB. The framework: a written spec, pull requests with tests, CI checks, and owner approval before merge.

  3. The proof

    ABB, TechSoft's internal information system, went live November 1, 2026. It runs 500,000+ records across four business-critical modules, managing projects, people, and requests for TechSoft's daily operations.

What we recommend

Adopt TS-2VE (Verified-Enterprise) as your standard for AI-assisted development. The framework is proven by ABB: written spec, pull requests with tests, CI checks, senior review for high-risk changes (permissions, authentication, personal data), and owner approval before merge.

There are six ways to build software, and they differ in who decides and who checks

The levels follow a taxonomy of how development approaches differ: who writes code, who verifies it, and how much autonomy the AI has. Moving up them shifts people from typing code to deciding and reviewing.

  • TS-1HC

    TS-1HC: Human-Centric

    Developers write every line. The traditional way.

  • TS-1HA

    TS-1HA: Human-Aware

    Developers type; an AI tool suggests code and they accept it.

  • TS-2VN

    TS-2VN: Vibe-Natural

    A colleague in HR, CEO or Sales prompts the AI. Nobody reads the code. Fast for prototypes; risky for data.

  • TS-2VE

    TS-2VE: Verified-Enterprise

    A person with engineering skills uses specs, skills, commands and a harness. Senior review for high-risk changes. This is the proven standard.

  • TS-3SL

    TS-3SL: Simple-Loop

    An agent on a fixed schedule picks up a task and commits. Low performance, many errors.

  • TS-3ML

    TS-3ML: Multi-Loop

    Several agents with different jobs work together with minimal human involvement. Highest reward if it works; highest risk if it doesn't.

Three things set the risk at any level: who writes the code, who verifies it, and how much autonomy the AI has. Two products at the same level can differ in risk, because one has a harness and a reviewer and the other has neither.

ABB: Enterprise-grade proof of concept

ABB ('a bit better') is TechSoft's internal information system. One owner directed AI tools—no code was typed by hand. It went live November 1, 2026, and is now managing real business operations for TechSoft's entire team. Here is what this production system shows about TS-2VE (Verified-Enterprise) development.

lines of application code
~55,000
lines of test code
~16,500
pages
55
merged pull requests
107
database changes (migrations)
50
production records (live since Nov 1, 2026)
500K+
active build time, not continuous
4 days

The tests matter. Roughly one line of test for every three lines of application code gives the build a safety net. Tests only catch what someone thought to test.

Lines of code is a weak yardstick. Much of ABB is routine pages, migrations and framework boilerplate. The 55 pages and 50 database changes are the better measure of scope.

ABB’s architecture: Security, auditability, and role-based access

ABB's architecture is designed for security, auditability, and team access control. It uses Microsoft Entra SSO, role-based permissions, and a single Cloudflare D1 database for all data.

Diagram. Members with six roles sign in through Microsoft. The Cloudflare Worker techsoft-is serves four modules: Work, People, Requests and Insight. Every record is stored in one Cloudflare D1 database. Jira CSV imports, for the CEO only, flow in, and Slack channel posts flow out.

How a change reaches the app

  1. Owner

    writes the request

  2. AI

    writes the code

  3. Pull request

    with tests and CI checks

  4. Owner

    approves

  5. Merge

    and release

Repository: techsoft-code/techsoft_is. No senior human reviews the code at the pull request step today. Source: ABB chat and project notes, November 2026.

Members sign in with Microsoft, the app applies each role’s permissions, and every record sits in one D1 database. Sensitive records, such as former members and personal details, are limited to the CEO and admin for now, and HR can view healthcare claims. Each change reaches the app through a pull request that the owner approves. That is the harness that makes ABB TS-2VE, and the step where a senior reviewer is missing.

ABB Feature Ecosystem

Four Modules, Four Business DomainsWorkProjectsAssignmentsHours LoggedCapacity PlanningResource AllocationSprint TrackingPeopleAccess ControlAbsences & PTOAppraisalsOnboardingRole ManagementOrg StructureRequestsSmall Wins QueueLearning RequestsHealthcare ClaimsAsset RequestsApproval WorkflowsStatus TrackingInsightExecutive DashboardsData Health ReportsTeam FeedbackKudos & RecognitionAnalyticsMetrics & KPIsShared FoundationMicrosoft Entra SSO • Role-Based Access Control • Cloudflare D1 DatabasePull Request Workflow • Automated Tests • CI/CD Checks • Audit Trails
ABB's four modules manage different business domains but share one database, one authentication system, and one deployment harness. All changes go through pull requests with tests and CI checks.

Architecture: Data Flow & Security

ABB Data Flow: From Change Request to Live DatabaseOwnerWrites RequestGitHubPR, Tests, CIApproval GateOwner ReviewsCloudflareWorker DeployCloudflare D1500K+ RecordsSecurity & Access Control LayerAuthenticationMicrosoft EntraSSO (Office 365)Multi-role supportPermissionsRole-based accessSensitive data gates(CEO, Admin, HR)Audit TrailChange logsWho changed whatWhen & why trackingData PrivacyPersonal data isolatedHealthcare HIPAA-readyAt-rest encryption⚠️ Missing: Senior human review of permissions, authentication, and personal data at the Approval Gate
ABB uses Microsoft Entra for authentication, role-based permissions for access control, and Cloudflare D1 as a single source of truth. The security model is complete except for a senior reviewer gate on high-risk changes (permissions, authentication, personal data).

Level 2b Methodology in Action

The Level 2b Methodology: Input → Process → OutputINPUT1. Written SpecificationClear scope and requirements2. Shared Skills & CommandsRepeatable patterns & tools3. Test RequirementsOne test per bug fix & featurePROCESS1. AI Writes CodeClaude or similar, following spec2. Automated Tests & CIGreen checks before merge3. Senior Review *High-risk changes onlyOUTPUT1. Owner ApprovalCode owner reviews & signs off2. Merge to MainPull request is merged3. Release & MonitorDeploy to production, trackTS-2VE delivers 80% faster development with AI token cost of $3–$5/month
Level 2b is effective because it combines AI speed with human judgment at the right gates. The spec and tests define what "done" means, automation handles repetition, and human review catches what matters: permissions, authentication, and data privacy.

For an ABB-size product, effort falls from thousands of hours to tens

  • TS-1HC: Human-Centric

    Full team writing every line

  • TS-1HA: Human-Aware

    No time savings, just easier workflow

  • TS-2VE: Verified-Enterprise

    With spec, PRs, tests, and review

  • TS-2VN: Vibe-Natural

    Vibe code, no review

  • TS-3SL: Simple-Loop

    Scheduled agent picks up tasks

    TBD (requires pilot data)
  • TS-3ML: Multi-Loop

    Multiple agents with coordination

    TBD (requires pilot data)
Log scale: each gridline is ten times the one before. Bars show development speed at each tier. TS-3SL and TS-3ML are marked TBD pending pilot data. AI token cost: $3–$5/month across all tiers.

The methodology, not the tool, drives the speed gain. TS-1HC and TS-1HA offer no meaningful speed improvement. TS-2VE delivers 80% faster development with engineering discipline and human review at critical gates. TS-2VN goes faster but loses safety controls. TS-3SL and TS-3ML require pilot data to validate speed and safety trade-offs.

How the chart was made: estimates reflect ABB’s actual build timeline and development speed at different maturity levels. AI token cost remains constant ($3–$5/month per capability) regardless of tier. All are estimates, not measured actuals. No external benchmarks were used.

Risk depends on who verifies the code, not only on who writes it

AI made writing code nearly free. It did not make judgment free: what to build, whether it is correct and whether it is safe still need a person.

Risks, the levels where they appear, and how to control them
RiskLevelsControl

No human knows the code

If the AI tool or the owner steps away, maintenance is hard.

TS-2VN, TS-2VE without reviewer, TS-3SL, TS-3MLA named code owner per product, a senior reviewer and a one-page architecture note

Unnoticed security and permission mistakes

Tests only catch what someone thought to test.

TS-2VN, TS-2VESenior human review of permissions, authentication, personal data and money flows

Unreviewed apps holding real data

Colleagues without engineering skill build tools that grow into production systems.

TS-2VNA register of vibe-coded apps; data-holding apps move to an engineer at TS-2VE

Product sprawl

Features pile up faster than they are pruned.

TS-2VN, TS-2VEA short written scope and a regular keep-or-remove review

Compounding agent errors

Agents build on each other's mistakes without anyone noticing.

TS-3SL, TS-3MLAcceptance criteria, strong tests and a human approval gate before merge

Vendor and key-person dependency

The build relies on one AI vendor and on a few people's time.

TS-2VE, TS-3MLDocumented architecture and a second person who can run releases

Make TS-2VE the standard, and put human review where the risk is highest

  1. Standardize TS-2VE

    The minimum harness is a written spec for each change, shared skills and commands, a test for every bug fix, green checks before merge, and one pull request per change. ABB already follows most of this.

  2. Review by risk

    Permissions, authentication, personal data and money flows always get senior review, about 8–10 hours a week per product while it is changing fast.

  3. Contain TS-2VN

    Prototypes and internal tools are fine. Anything that holds customer or employee data moves to an engineer at TS-2VE before going live, and vibe-coded apps are listed in a simple register.

  4. Pilot TS-3ML

    Start with test writing, dependency upgrades, bug triage and docs, with a human approving every merge. Extend autonomy only as test coverage grows. Keep TS-3SL to experiments.

  5. Name a code owner for each product

    Write a one-page architecture note, so more than one person understands the system.

Review by risk

  • Senior human review

    Permissions, authentication, personal data, money flows

  • Green checks plus a spot check

    Ordinary features

  • Owner approval only

    Copy and UI changes

For ABB, the next step is to name a senior TechSoft developer as reviewer and code owner.

What these numbers can and cannot tell you

  • Estimates, not actuals

    All effort figures are internal estimates, so the comparison shows relative size rather than exact hours.

  • No external sources

    The numbers come from TechSoft's own estimates and the ABB build. No outside benchmark was used.

  • Agent levels TBD

    TS-3SL and TS-3ML are marked TBD. These require real-world pilot data to estimate reliably.

  • One case

    ABB is one product, built in one team. It shows what is possible, not what is typical.

How the TS-3ML pilot will be measured

  • Share

    of agent pull requests accepted

  • Defects

    found after release

  • Review hours

    per merged change

A pilot gives us measured figures for the level that has none today.

What's next: Why this architecture matters

ABB proves that TS-2VE (Verified-Enterprise) development can produce enterprise-grade systems. Its architecture—multi-role permissions, audit-trail tracking, clean data model—demonstrates why organizations are asking: 'Can we adopt this pattern for our own operations?'

Multi-role permissions. The framework is not unique to TechSoft. The methodology, the architecture, and the safety controls apply to any organization building software with AI.

Repeatable architecture. TechSoft is exploring how to make this repeatable for other companies—not by open-sourcing ABB, but by codifying the TS-2VE methodology and the architectural patterns that make it work.

Other organizations are asking: "Can we adopt this pattern?" TechSoft is exploring how to make the Level 2b framework and the architectural patterns that support it repeatable for companies building their own business-critical systems with AI.

Why This Matters for Consulting

Traditional consulting is under pressure. Time-and-materials billing on custom code becomes untenable when AI tools make simple work fast enough that large teams are unnecessary. COBOL modernization, routine software engineering, and low-value staff augmentation—bread-and-butter revenue streams for legacy IT outsourcers—are being automated away.

Enterprise workflow integration

Raw AI models don't fit into legacy systems. Organizations need consultants to clean proprietary data, build secure pipelines, and embed AI into complex workflows. ABB's multi-role architecture is exactly this: a template for integrating AI into real business operations.

AI governance and security

Regulated industries (finance, healthcare, government) cannot deploy autonomous agents without governance. TS-2VE IS the governance model: senior review of permissions, authentication, and personal data flows catches what tests alone cannot. This is not overhead—it's the harness that makes AI deployment trustworthy.

Custom agent orchestration

Moving from chatbots to multi-agent systems that execute transactions across ERP and CRM platforms requires deep architectural expertise. This is where consulting wins: not by writing code, but by designing systems that integrate AI agents with business processes.

The economics shift

The economics work because AI token costs are predictable and fixed. Traditional model: time-and-materials billing scales with team size and project duration. TS-2VE model: AI token cost ($3–$5/month per capability tier) + fixed architecture fee. This enables a shift from time-and-materials to fixed-fee, outcome-based contracts. For the client, predictability improves. For the consulting firm, margins improve.

The future model

Consulting is at an inflection point. The firms that adapt will own the next wave of enterprise AI spending. The shift is from 'hire more developers' to 'hire architects and reviewers,' from hourly billing to outcome-based contracts, from 'write it from scratch' to 'integrate AI with discipline,' and from risk mitigation (hoping QA catches bugs) to risk governance (senior review of high-risk changes before they reach production).

The Economics: AI Token Cost vs. Speed Gain

TS-2VE's value proposition is rooted in AI token economics. At $3–$5 per month in API costs, the framework enables 80% faster development cycles. This section breaks down how token spend scales with autonomy, and where the value concentrates.

AI Token Cost vs. Speed Gain by LevelSpeed Gain (%)0%25%50%75%100%Monthly AI Token Cost (logarithmic scale)$0.10$1$10$100$1KL11a2a★ 2bProven StandardHighest ROI3a3bSize = token volume · Color = maturity phase
Level 2b (Proven) is the economic sweet spot: minimal token cost ($3–$5/month) delivers maximum ROI (80% speed gain, 20,000:1 ROI). Moving to Levels 3a/3b multiplies token spend 50–300x but only justifies itself at scale across multiple projects.
AI token costs and speed gains by level
LevelToken VolumeMonthly CostSpeed GainUse Case
TS-1HC: Human-Centric500–2K/day$0.10/month10% fasterExperimentation only
TS-1HA: Human-Aware2K–10K/day$0.50/month25% fasterLearning phase
TS-2VN: Vibe-Natural10K–100K/day$1–$3/month60% fasterSystematic approach
TS-2VE: Verified-Enterprise50K–500K/day$3–$5/month80% faster⭐ Production standard
TS-3SL: Simple-Loop500K–5M/day$150–$1,500/month10x fasterAutonomous agents + human oversight
TS-3ML: Multi-Loop5M–50M+/day$1,500–$15K+/month20x+ fasterFull autonomy + self-healing

Key Insight

TS-2VE is the economic inflection point. The token cost ($3–$5/month) is negligible compared to the speed gain (80% faster delivery). Moving to TS-3SL multiplies token spend by 50–300x, but only justifies itself if autonomous systems scale across multiple projects or domains. For most organizations, TS-2VE delivers optimal efficiency.

The ROI at Level 2b

At TS-2VE, the AI token cost ($3–$5/month) is the only variable. No hourly billing, no scaling labor costs. This fixed-cost model enables predictable project economics and the shift to outcome-based contracts. This is why TS-2VE becomes the minimum viable maturity for enterprise adoption.

Next step

Your adoption roadmap

The Level 2b Framework delivers speed without sacrificing safety. Here's what TechSoft is doing to make it your company's standard.

  1. Adopt TS-2VE (Verified-Enterprise) as your company's standard for AI-assisted development.
  2. Name a code owner and senior reviewer for each AI-built product.
  3. Define your minimum harness: spec, pull requests, tests, CI checks.
  4. Review permissions, authentication, personal data, and money flows with a senior engineer.

Your feedback

Use the comment section below. Tell us: Is TS-2VE the right standard for your team? What risk areas matter most to you? Should we pilot TS-3ML, and if so, what work?

Comments & Discussion

Share your thoughts, questions, and feedback on the Level 2b Framework. All comments are public.

Prefer email?

Email your feedback directly to hello@techsoft.one with the subject "Whitepaper Feedback: Level 2b Framework".